CN101408945A - Method for sorting radar two-dimension image base on multi-dimension geometric analysis - Google Patents
Method for sorting radar two-dimension image base on multi-dimension geometric analysis Download PDFInfo
- Publication number
- CN101408945A CN101408945A CNA2008102324508A CN200810232450A CN101408945A CN 101408945 A CN101408945 A CN 101408945A CN A2008102324508 A CNA2008102324508 A CN A2008102324508A CN 200810232450 A CN200810232450 A CN 200810232450A CN 101408945 A CN101408945 A CN 101408945A
- Authority
- CN
- China
- Prior art keywords
- feature
- energy
- energy feature
- radar
- image
- 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
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000004458 analytical method Methods 0.000 title claims abstract description 14
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 20
- 239000000284 extract Substances 0.000 claims abstract description 11
- 230000004927 fusion Effects 0.000 claims abstract description 9
- 238000010606 normalization Methods 0.000 claims abstract description 9
- 239000004744 fabric Substances 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 230000008676 import Effects 0.000 claims description 2
- 230000007812 deficiency Effects 0.000 abstract description 5
- 238000012545 processing Methods 0.000 abstract description 2
- 238000002474 experimental method Methods 0.000 description 34
- 238000012549 training Methods 0.000 description 17
- 238000012360 testing method Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 7
- 230000009466 transformation Effects 0.000 description 5
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
本发明公开了一种基于多尺度几何分析的雷达二维像分类方法，它属于图像处理技术领域，主要克服现有方法不能有效的表示雷达二维像的不足。具体步骤为：(1)输入样本图像集，并对样本图像集的每一幅图像进行归一化；(2)对归一化后每一个样本图像进行三层小波分解和Contourlet分解，得到一个样本图像分别对应的10个小波分解子带和13个Contourlet分解子带；(3)对得到的分解子带提取能量特征，并利用特征融合的方法对这些能量特征进行合并；(4)选择支撑矢量机算法对合并后的特征进行分类。本发明具有较好的分类准确率以及较低的复杂度，可用于雷达二维像、纹理图像的分类及SAR图像中桥梁目标识别。
The invention discloses a radar two-dimensional image classification method based on multi-scale geometric analysis, which belongs to the technical field of image processing and mainly overcomes the deficiency that the existing method cannot effectively represent the radar two-dimensional image. The specific steps are: (1) Input the sample image set, and normalize each image in the sample image set; (2) Perform three-layer wavelet decomposition and Contourlet decomposition on each sample image after normalization to obtain a The 10 wavelet decomposition subbands and 13 Contourlet decomposition subbands corresponding to the sample image respectively; (3) extract the energy features from the obtained decomposition subbands, and use the method of feature fusion to merge these energy features; (4) select the support The vector machine algorithm classifies the merged features. The invention has better classification accuracy and lower complexity, and can be used for classification of radar two-dimensional images and texture images and bridge target recognition in SAR images.
Description
Technical field
The present invention relates to technical field of image processing, a kind of sorting technique of image particularly can be used for bridge Target Recognition in the classification of radar two-dimensional image, texture image and the SAR image.
Background technology
Comprised many linear informations in the general radar two-dimensional image, these information in image any position of any direction and different scale on all may show.Small echo can only provide limited several directions, therefore can not fully excavate the directional information in the radar two-dimensional image.And the good assistant that multiple dimensioned how much instruments are dealt with problems just has great potential when handling linear feature.Small echo has optimum characterization when approaching the objective function with one dimension odd opposite sex, promptly put singularity, yet under the high dimensional data situation, small echo can not optimumly represent that some has the function of geometric properties.For example the two-dimentional separable tensor product small echo that is made of the one dimension small echo has only limited direction, it between its Support square, but a lot of geometries in the two dimensional image have directivity, thereby it can not handle this class image well, and small echo can't provide a description the coefficient at direction and edge.Many scholars have proposed the deficiency that the multi-scale geometric analysis theory overcomes small echo, as methods such as ridge ripple, curve ripple, profile ripples.People such as M.N.Do and Martin Vetterli have proposed a kind of new multi-scale geometric analysis instrument-Contourlet in 2002.The profile ripple is many resolutions, graphical representation method local, direction, can represent to comprise the image that enriches profile and texture effectively.
Aspect the classification of radar two-dimensional image, feature extraction is necessary and a crucial step.The quality of feature extraction is directly concerning the correctness and the validity of classification results.How to select the method that effectively to extract characteristics of image very important for follow-up classification and identification, general feature extracting method has energy feature, variance, average, Hu square etc., yet certain feature of single selection can not effectively be represented image to be classified, can't embody the validity of characteristic of division.
Summary of the invention
The objective of the invention is to overcome the deficiency of above-mentioned prior art, proposed a kind of sorting technique,, realize effectively expression radar two-dimensional image to improve the validity of characteristic of division based on multi-scale geometric analysis.
Realize that the object of the invention technical scheme comprises following process:
(1) imports the sample image collection, and each width of cloth image of sample image collection is carried out normalization;
(2) each sample image after the normalization is carried out three layers of wavelet decomposition and Contourlet decomposition, obtain a sample image corresponding respectively 10 wavelet decomposition subbands and 13 Contourlet and decompose subbands;
(3) each that obtains is decomposed subband and extract energy feature, and utilize the method for Feature Fusion that these energy features are merged;
(4) select the energy feature after the supporting vector machine algorithm is combined to classify.
Above-mentioned radar two-dimensional image sorting technique based on multi-scale geometric analysis, wherein the described method of Feature Fusion of utilizing of step (3) merges energy feature, and detailed process is as follows:
3a) with wavelet decomposition sub belt energy feature w={w1, w2 ..., w10} and Contourlet decompose sub belt energy feature c={c1, c2 ..., c13} is arranged side by side, obtain corresponding 23 subbands energy feature w1, w2 ..., w10, c1, c2 ..., c13};
3b) to the subband energy feature w1, w2 ..., w10, c1, c2 ..., c13} selects, the energy feature after the selection then is the energy feature after merging.
Above-mentioned energy feature merging method, wherein step 3b) described to the subband energy feature w1, w2 ..., w10, c1, c2 ..., c13} selects, detailed process is as follows:
3b.1) with the energy feature of subband w1, w2 ..., w10, c1, c2 ..., c13}, the series arrangement descending according to eigenwert be, and cw1, cw2 ..., cw23};
3b.2) give up 8 features that are arranged in the back, stay preceding 15 energy features as the energy feature after merging.The present invention compared with prior art has following advantage:
1. the present invention is owing to adopted the Feature Fusion technology, wavelet decomposition and Contourlet are decomposed the energy feature merging of extracting the back, make that the energy feature after merging is more effective, overcome the deficiency of feature when expression radar two-dimensional image that wavelet transformation extracts, can access better classification results.
2. the present invention has given up the part energy feature because the sub belt energy feature is side by side selected, and has reduced the dimension of energy feature, has further reduced the complexity of the inventive method, has improved performance.
Test experiments shows that classification accuracy of the present invention is 88.21%, is better than the classification accuracy 83.53% of wavelet method and the classification accuracy 84.72% of Contourlet method.
Description of drawings
Fig. 1 is realization flow figure of the present invention;
Fig. 2 is that the present invention tests 4 class aircraft sample synoptic diagram that adopt;
Fig. 3 be the present invention test two and the experiment the three 10 class aircraft sample synoptic diagram that adopted;
Fig. 4 is that the present invention tests the three 2 class naval vessel sample synoptic diagram that adopt.
Embodiment
With reference to Fig. 1, step of the present invention is as follows:
Suppose that sample image collection number to be classified is the N width of cloth, be designated as (x
1, x
2..., x
N); Because what adopt is the sorting algorithm of supervised learning, i.e. supporting vector machine algorithm is so need be designated as (x as training sample image by the picked at random M width of cloth from N width of cloth image
1, x
2..., x
M); Image set (x
1, x
2..., x
N) pixel between 0～255 normalizing to 0～1, be designated as (z
1, z
2..., z
N).According to the proof of Donoho, the image after the normalization can extract more effectively feature.
Step 2 is carried out three layers of wavelet decomposition and Contourlet decomposition to each sample image after the normalization, obtains a sample image corresponding respectively 10 wavelet decomposition subbands and 13 Contourlet and decomposes.
To the sample image collection (z after the normalization
1, z
2..., z
N) carry out three layers of wavelet transformation and Contourlet conversion respectively.Sample image is concentrated j sample image z
j, carry out three layers of wavelet transformation, can obtain 10 and decompose subband; Be to this sample image z equally
jCarry out three layers of Contourlet conversion, set the maximum detail layer and get 8 direction subbands, inferior maximum detail layer is got 4 direction subbands, adds 1 approximate subband, draws the subband after corresponding 13 Contourlet decompose.
Step 3 is extracted energy feature to the decomposition subband that obtains, and the method for utilizing Feature Fusion merges the feature that wavelet decomposition subband and Contourlet decompose subband.
3a. extract the energy feature that each decomposes subband.
For the subband after decomposing, need to extract the feature of each subband, these features comprise energy feature, variance, average, Hu square, through a large amount of experiments relatively, energy feature is the most effective in these several features, so the present invention extracts energy feature.Extracting method adopts L
1Norm energy norm method:
Wherein, P * Q is for extracting the subband size of feature, and i represents the line number of the feature subband that extracts, and j represents the columns of the feature subband that extracts, coef (i, j) by the coefficient value that the capable j of i is listed as in the extraction feature subband.
3b. utilize the method for Feature Fusion that energy feature is merged.
The Feature Fusion method that adopts among the present invention is that feature is arranged side by side, increases the number of feature, needs afterwards feature is selected, and realizes the purpose of dimensionality reduction.The present invention is for simple and practical, with wavelet decomposition sub belt energy feature w={w1, and w2 ..., w10} and Contourlet decompose sub belt energy feature c={c1, c2 ..., c13} is arranged side by side, obtains energy feature { w1, the w2 of corresponding 23 subbands, ..., w10, c1, c2 ..., c13}, then to the subband energy feature w1, and w2 ..., w10, c1, c2, ..., c13} selects, and the series arrangement descending according to eigenwert is, cw1, cw2 ..., cw23} gives up 8 features that are arranged in the back, stays preceding 15 energy features as the energy feature after merging.
In the test experiments of reality, the energy feature that extracts from M width of cloth training sample image is corresponding with training sample image, is designated as training characteristics (f
1, f
2..., f
M), all N width of cloth treat that the energy feature that the classification samples image extracts is designated as characteristic of division (f
1, f
2..., f
N).
Step 4 selects the energy feature after the supporting vector machine algorithm is combined to classify.
The sorting technique of selecting among the present invention is a kind of sorting algorithm that supervision is arranged, it is the supporting vector machine algorithm, be abbreviated as SVM, this algorithm to the process that image carries out classification and Detection is: 1. with the input as sorting algorithm of the training characteristics that obtains, corresponding with it sorting algorithm output result is desirable output result, thus the training sorting algorithm; 2. in the sorting algorithm that the input of next step characteristic of division that obtains has been trained, the output of sorting algorithm is last classification results.
According to the step explanation of SVM algorithm, the present invention is as follows to the detailed process of energy feature classification:
4a. with the training characteristics (f described in the step 3
1, f
2..., f
M), and one group of desirable output result (y of this feature correspondence
1, y
2..., y
M), as the training set of SVM;
4b. with described training set substitution formula
{ k represents training characteristics (f in the formula for α, b} to obtain optimum solution
1, f
2..., f
M) ordinal number, α represents the weights of SVM, b represents the inclined to one side value of SVM;
4c. with optimum solution α, b} construct the corresponding relation of SVM input and output:
Output
SVM=sgn (α * defeated
SVM+ b) (2)
Sgn () is-symbol function wherein,
4d. with the characteristic of division (f described in the step 3
1, f
2..., f
N) input in the replacement formula (2)
SVM, the then output in the formula (2)
SVMThe value that obtains is exactly the classification results of required radar two-dimensional image.
Performance of the present invention can be tested by the classification experiments of following three radar two-dimensional images, and following experimental data has provided the average result of 20 times, 50 times and the 100 times independent operatings that circulate respectively.All experimental results all are under Windows XP service condition, and CPU obtains in MATLAB 7.0 environment for the IV 2.4GHz that runs quickly.
Experiment one: based on the radar two-dimensional image middle and small scale aircraft sample classification experiment of multi-scale geometric analysis.
This experiment is used for checking the influence of the inventive method to the classification performance of radar two-dimensional image middle and small scale aircraft sample.
For the validity of the energy feature after the present invention merges is described, the method with wavelet transformation and Contourlet conversion compares respectively.
The data set of experiment is radar two-dimensional image storehouse, and sample size is 128 * 128.230 samples have been selected in the experiment altogether, 4 class aircraft images have been comprised, the all corresponding certain fragmentary sample of every class, 9 fragmentary samples are arranged in this experiment, and when training SVM, 20 samples of picked at random are as training sample from 230 all samples, and Fig. 2 is the experiment one 4 class aircraft sample synoptic diagram that adopted.Provided the average result of three kinds of methods in the table 1, comprised test accuracy rate, repeatedly a best classification results in service and the average wrong sample number that divides among the result at 20 times, 50 times and 100 times independent operatings of circulation.
Table 1 is based on the radar two-dimensional image middle and small scale aircraft sample reason classification experiments of multi-scale geometric analysis
Data in the table 1 show, in 20 times result of circulation, the classification accuracy that sorting technique of the present invention obtains is better than other two methods, yet in the result of circulation 50 times and 100 times, result of the present invention is slightly poorer than the Contourlet method, be because specimen types is few, and the sample of selecting in this experiment is suitable for Contourlet method condition, so in experiment subsequently, increased sample number and specimen types number.
Experiment two: fairly large aircraft sample classification experiment in the radar two-dimensional image based on multi-scale geometric analysis.
This experiment is used for checking the method for mentioning among the present invention to aircraft sample classification Effect on Performance fairly large in the radar two-dimensional image, and by adopting the influence of feature selecting technology to algorithm performance among comparative descriptions the present invention of working time.
The data set of experiment is radar two-dimensional image storehouse, and size is 128 * 128.800 samples have been selected in the experiment altogether, 10 class aircraft images have been comprised, 80 samples of every class, and all corresponding certain fragmentary sample, 32 fragmentary samples are arranged in this experiment, and when training SVM, 100 samples of picked at random are as training sample from 800 all samples, and Fig. 3 is the experiment two 10 class aircraft sample synoptic diagram that adopted.Provided the average result of three kinds of methods in the table 2, comprised test accuracy rate, repeatedly a best classification results in service and the average wrong sample number that divides among the result at 20 times, 50 times and 100 times independent operatings of circulation.
Fairly large aircraft sample reason classification experiments in the radar two-dimensional image of table 2 based on multi-scale geometric analysis
Data by table 2 show that the classification accuracy that the present invention obtains is better than other two methods, under the condition that fragmentary sample exists, still can reach 90% test accuracy rate.
For feature selecting operation that the present invention the adopts influence to the inventive method is described, provided in the table 3 and taked feature selecting and do not take the influence of feature selecting working time of the present invention.
Table 3 comparison working time (unit: second)
| Take feature selecting | Do not take feature selecting |
| 51.68 | 72.69 |
By comparing data explanation working time of table 3, the present invention has given up the part energy feature because the sub belt energy feature is side by side selected, and has further reduced complexity, has accelerated working time.
Experiment three: fairly large mixing aircraft and naval vessel sample classification experiment in the radar two-dimensional image based on multi-scale geometric analysis.
This experiment is used for checking method of the present invention to fairly large mixing aircraft in the radar two-dimensional image and naval vessel sample classification Effect on Performance.
The data set of experiment is radar two-dimensional image storehouse, and size is 128 * 128.Selected 1000 samples in the experiment altogether, comprised 10 class aircraft images, 80 samples of every class, 800 samples altogether; 2 class ship images, 100 samples of every class, have 200 samples altogether, and all corresponding certain fragmentary sample, 40 fragmentary samples are arranged in this experiment, and when training SVM, 120 samples of picked at random are as training sample from 1000 all samples, Fig. 3 is the experiment three 10 class aircraft sample synoptic diagram that adopted, and Fig. 4 is the experiment three 2 class naval vessel sample synoptic diagram that adopted.Provided the average result of three kinds of methods in the table 4, comprised test accuracy rate, repeatedly a best classification results in service and the average wrong sample number that divides among the result at 20 times, 50 times and 100 times independent operatings of circulation.
Fairly large mixing aircraft and naval vessel sample classification experiment in the radar two-dimensional image of table 4 based on multi-scale geometric analysis
As can be seen from Table 4, the present invention can effectively classify to the image that has mixed aircraft and naval vessel, and its classification results is better than other two kinds of methods, and classification accuracy is preferably arranged.
Above-mentioned three experiments all as can be seen, Feature Fusion technology and feature selection approach that the present invention adopts can effectively overcome the deficiency that wavelet transformation is represented the radar two-dimensional image, have improved classification accuracy.In addition, three experiments among the present invention all adopt about 10% in all samples as training sample, illustrate that method of the present invention is suitable for condition of small sample, and this is highly significant for the radar image of handling other.
Claims (3)
1. radar two-dimensional image sorting technique based on multi-scale geometric analysis comprises following process:
(1) imports the sample image collection, and each width of cloth image of sample image collection is carried out normalization;
(2) each sample image after the normalization is carried out three layers of wavelet decomposition and Contourlet decomposition, obtain a sample image corresponding respectively 10 wavelet decomposition subbands and 13 Contourlet and decompose subbands;
(3) each that obtains is decomposed subband and extract energy feature, and utilize the method for Feature Fusion that these energy features are merged;
(4) select the energy feature after the supporting vector machine algorithm is combined to classify.
2. radar two-dimensional image sorting technique according to claim 1, wherein the described method of Feature Fusion of utilizing of step (3) merges energy feature, and detailed process is as follows:
3a) with wavelet decomposition sub belt energy feature w={w1, w2 ..., w10} and Contourlet decompose sub belt energy feature c={c1, c2 ..., c13} is arranged side by side, obtain corresponding 23 subbands energy feature w1, w2 ..., w10, c1, c2 ..., c13};
3b) to the subband energy feature w1, w2 ..., w10, c1, c2 ..., c13} selects, the energy feature after the selection then is the energy feature after merging.
3. energy merging method according to claim 2, wherein step 3b) described to the subband energy feature w1, w2 ..., w10, c1, c2 ..., c13} selects, detailed process is as follows:
3b.1) with the energy feature of subband w1, w2 ..., w10, c1, c2 ..., c13}, the series arrangement descending according to eigenwert be, and cw1, cw2 ..., cw23};
3b.2) give up 8 features that are arranged in the back, stay preceding 15 energy features as the energy feature after merging.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008102324508A CN101408945B (en) | 2008-11-28 | 2008-11-28 | Classification Method of Radar Two-Dimensional Image Based on Multi-scale Geometric Analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008102324508A CN101408945B (en) | 2008-11-28 | 2008-11-28 | Classification Method of Radar Two-Dimensional Image Based on Multi-scale Geometric Analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101408945A true CN101408945A (en) | 2009-04-15 |
| CN101408945B CN101408945B (en) | 2011-04-27 |
Family
ID=40571953
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2008102324508A Expired - Fee Related CN101408945B (en) | 2008-11-28 | 2008-11-28 | Classification Method of Radar Two-Dimensional Image Based on Multi-scale Geometric Analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101408945B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101799873B (en) * | 2010-01-28 | 2011-10-19 | 哈尔滨工业大学 | Multi-group image supervised classification method based on empirical mode decomposition |
| CN102270295A (en) * | 2011-07-01 | 2011-12-07 | 西安电子科技大学 | Fast Bridge Detection Method in SAR Image |
| CN101634706B (en) * | 2009-08-19 | 2012-01-04 | 西安电子科技大学 | Method for automatically detecting bridge target in high-resolution SAR images |
| CN103792523A (en) * | 2014-03-05 | 2014-05-14 | 西安科技大学 | UHF waveband multi-channel radar radial speed detection method based on tensor product |
| CN105930788A (en) * | 2016-04-18 | 2016-09-07 | 太原理工大学 | Non-downsampling contour wave and PCA (principal component analysis) combining human face recognition method |
| CN106485278A (en) * | 2016-10-13 | 2017-03-08 | 河南科技大学 | A kind of image texture sorting technique based on shearing wave and gauss hybrid models |
-
2008
- 2008-11-28 CN CN2008102324508A patent/CN101408945B/en not_active Expired - Fee Related
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101634706B (en) * | 2009-08-19 | 2012-01-04 | 西安电子科技大学 | Method for automatically detecting bridge target in high-resolution SAR images |
| CN101799873B (en) * | 2010-01-28 | 2011-10-19 | 哈尔滨工业大学 | Multi-group image supervised classification method based on empirical mode decomposition |
| CN102270295A (en) * | 2011-07-01 | 2011-12-07 | 西安电子科技大学 | Fast Bridge Detection Method in SAR Image |
| CN103792523A (en) * | 2014-03-05 | 2014-05-14 | 西安科技大学 | UHF waveband multi-channel radar radial speed detection method based on tensor product |
| CN105930788A (en) * | 2016-04-18 | 2016-09-07 | 太原理工大学 | Non-downsampling contour wave and PCA (principal component analysis) combining human face recognition method |
| CN106485278A (en) * | 2016-10-13 | 2017-03-08 | 河南科技大学 | A kind of image texture sorting technique based on shearing wave and gauss hybrid models |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101408945B (en) | 2011-04-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jia et al. | Gabor feature-based collaborative representation for hyperspectral imagery classification | |
| Yadav et al. | Multiresolution local binary pattern variants based texture feature extraction techniques for efficient classification of microscopic images of hardwood species | |
| Sirmacek et al. | Urban area detection using local feature points and spatial voting | |
| Yang et al. | CNN with spatio-temporal information for fast suspicious object detection and recognition in THz security images | |
| CN101408945B (en) | Classification Method of Radar Two-Dimensional Image Based on Multi-scale Geometric Analysis | |
| Prasad et al. | Plant leaf species identification using curvelet transform | |
| Jia et al. | A two-stage feature selection framework for hyperspectral image classification using few labeled samples | |
| CN105913081B (en) | SAR image classification method based on improved PCAnet | |
| CN104778482B (en) | The hyperspectral image classification method that dimension about subtracts is cut based on the semi-supervised scale of tensor | |
| CN105989368A (en) | Target detection method and apparatus, and mobile terminal | |
| CN101763514A (en) | Image segmentation method based on characteristic importance sorting spectral clustering | |
| CN105678788A (en) | Fabric defect detection method based on HOG and low-rank decomposition | |
| CN101551864A (en) | Image classification method based on feature correlation of frequency domain direction | |
| CN108764310A (en) | SAR target identification methods based on multiple dimensioned multiple features depth forest | |
| Valliammal et al. | A novel approach for plant leaf image segmentation using fuzzy clustering | |
| CN107273919A (en) | A kind of EO-1 hyperion unsupervised segmentation method that generic dictionary is constructed based on confidence level | |
| Rathore et al. | Classification of colon biopsy images based on novel structural features | |
| CN112381144A (en) | Heterogeneous deep network method for non-European and European domain space spectrum feature learning | |
| Ahmed et al. | Predicting skin cancer melanoma using stacked convolutional neural networks model | |
| Chen et al. | Comparative study on deep-learning-based leather surface defect identification | |
| Manikis et al. | Pollen grain classification using geometrical and textural features | |
| He et al. | Convolutional transformer-inspired autoencoder for hyperspectral anomaly detection | |
| Vishwakarma et al. | Generalized DCT and DWT hybridization based robust feature extraction for face recognition | |
| Li et al. | A simple scheme to amplify inter-class discrepancy for improving few-shot fine-grained image classification | |
| Yadav et al. | Analysis and classification of hardwood species based on Coiflet DWT feature extraction and WEKA workbench |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210407 Address after: 710076 room 104, block B2, software new town phase II, tianguba Road, Yuhua Street office, high tech Zone, Xi'an City, Shaanxi Province Patentee after: Discovery Turing Technology (Xi'an) Co.,Ltd. Address before: 710071 No. 2 Taibai Road, Shaanxi, Xi'an Patentee before: XIDIAN University |
|
| TR01 | Transfer of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20110427 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |