CN104751172B - The sorting technique of Polarimetric SAR Image based on denoising autocoding - Google Patents
The sorting technique of Polarimetric SAR Image based on denoising autocoding Download PDFInfo
- Publication number
- CN104751172B CN104751172B CN201510108639.6A CN201510108639A CN104751172B CN 104751172 B CN104751172 B CN 104751172B CN 201510108639 A CN201510108639 A CN 201510108639A CN 104751172 B CN104751172 B CN 104751172B
- Authority
- CN
- China
- Prior art keywords
- sar image
- networks
- polarimetric sar
- layer
- autocoding
- 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.)
- Active
Links
Landscapes
- Image Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of Classification of Polarimetric SAR Image methods based on denoising autocoding DA networks, mainly solve the problems, such as that complicated prior art extraction characteristic procedure, feature generalization ability difference and nicety of grading are low.Implementation step is：One Polarimetric SAR Image to be sorted of input option first extracts the primitive character and its neighborhood characteristics of the Polarimetric SAR Image；Then it takes the logarithm to primitive character and neighborhood characteristics processing, its noise is made to meet Gaussian Profile；Secondly the number of plies, each node layer number and the data noise of denoising autocoding DA networks and training denoising autocoding DA networks are determined；Then using trained denoising autocoding DA networks, classify to Polarimetric SAR Image to be sorted, obtain the classification results of Polarimetric SAR Image.The present invention simplifies the process of feature extraction, the generalization ability for improving feature and the nicety of grading to image, the Objects recognition available for Polarimetric SAR Image due to the use of denoising autocoding DA networks.
Description
Technical field
The invention belongs to technical field of image processing, further relate to the number in Classification of Polarimetric SAR Image technical field
According to feature extraction mode and depth network class etc., the Objects recognition available for Polarimetric SAR Image.
Background technology
Polarization SAR is a kind of high-resolution active-mode active microwave remote sensing imaging radar, has round-the-clock, round-the-clock, divides
Resolution is high, can survey the advantages that regarding imaging, can be applied to the numerous areas such as military, agricultural, navigation, geography monitoring.Compared with SAR,
What polarization SAR carried out is Polarimetry, can obtain the more rich information of target.In recent years, it is carried out using polarization SAR data
It is sorted in international remote sensing fields to be highly valued, it has also become the main direction of studying of image classification.
With the development of polarization SAR, the method for Classification of Polarimetric SAR Image is also to emerge in an endless stream.It is many unsupervised and have
The method of supervision makes polarization SAR classification accuracy rate be greatly improved.Unsupervised method is divided into two classes：One kind is using poly-
The mode of class, such as K mean values, Isodata；Another kind of is the non-coherent nature using data, as Cloude and Pottier is proposed
The distributive sorting (1997) based on entropy H and α angles, Lee etc. combined Freeman-Durden and decomposes and be distributed based on multiple Wishart
Maximum a posteriori probability classification (2004) etc.；These two kinds of methods are required to spend very big calculating cost, and accuracy is opposite
It is relatively low.The method for having supervision is mainly the following：ANN points of the artificial neural network that Heermann and Khazanie et al. are proposed
The sorting technique based on support vector machines that class method (1992), Burges and Vapnik are proposed, these have the method for supervision
Arithmetic speed and accuracy are substantially increased, but there is very strong dependence in the selection of feature, thus to classification results
Accuracy influence it is bigger.
Invention content
Present invention aims at the deficiencies for above-mentioned prior art, propose a kind of polarization based on denoising autocoding
SAR image sorting technique to be effectively retained the useful information of Polarimetric SAR Image, improves the nicety of grading of Polarimetric SAR Image.
The present invention technical thought be：It is extracted on the basis of initial data by depth network more advanced useful
Feature so that feature possesses stronger Generalization Capability, while has the energy for filtering out noise using denoising autocoding DA networks
Power, avoid using wave filter and caused by part useful information lose so that the more robust expression of data improves classification results
Accuracy.
According to above-mentioned thinking, technical step of the invention includes as follows：
(1) one Polarimetric SAR Image to be sorted of input option, extract the Polarimetric SAR Image primitive character and its
Neighborhood characteristics；
(2) it takes the logarithm to primitive character and neighborhood characteristics processing, its noise is made to meet Gaussian Profile；
(3) number of plies, each node layer number and the data noise of denoising autocoding DA networks are determined；
(4) training denoising autocoding DA networks
(4a), with reference to figure, determines the classification number with reference to atural object in figure, 10% ground is chosen in every class according to practical atural object
Substance markers are input to as training sample, and by the feature of training sample in the autocoding DA networks, successively covet
Greedy training；
(4b) finely tunes the structural parameters and noise parameter of denoising autocoding DA networks using back-propagation algorithm BP,
Obtain trained denoising autocoding DA networks；
(5) trained denoising autocoding DA networks are utilized, classifies to Polarimetric SAR Image to be sorted, obtains
The classification results of Polarimetric SAR Image.
Compared with prior art, the present invention it has the following advantages：
First, the present invention is due to transformation of taking the logarithm during polarization SAR data so that noise approximation Gaussian distributed,
Simplify noise model, it is easier to which classification learning algorithm filters out the noise of polarization SAR data；
Second, the present invention is directly trained with the primitive character and neighborhood characteristics of polarization SAR data input network, is retained
The useful information of polarization SAR, improves classification accuracy rate；
Third, the present invention add Gaussian noise to autocoding network, network are made to remove this as possible in learning process
It plants noise and obtains the characteristic crossed by a small amount of noise pollution, improve the accuracy of classification.
Description of the drawings
Fig. 1 is the realization flow chart of the present invention；
Fig. 2 is the tag along sort figure of Polarimetric SAR Image that present invention emulation uses；
Fig. 3 is the classification results figure to Polarimetric SAR Image to be sorted using existing depth belief network DBN；
Fig. 4 is the classification results figure for the Polarimetric SAR Image classified using existing autocoding network handles；
Fig. 5 is the classification results figure to Polarimetric SAR Image to be sorted with the present invention.
Specific embodiment
Technical solutions and effects of the present invention is described in further detail below in conjunction with attached drawing.
With reference to Fig. 1, specific implementation step of the invention is as follows：
The primitive character and its neighborhood characteristics of step 1, extraction Polarimetric SAR Image to be sorted.
One Polarimetric SAR Image to be sorted of (1a) input option decomposes the relevant of the Polarimetric SAR Image according to the following formula
Matrix：
Wherein T represents the coherence matrix of Polarimetric SAR Image, and i represents imaginary part unit,Polarimetric SAR Image
The value each put is the coherence matrix of a 3*3, and a represents the symmetrical factor of Polarimetric SAR Image, and c represents Polarimetric SAR Image
The configuration factor, d represent the local curvature of Polarimetric SAR Image, and h represents the directionality of Polarimetric SAR Image, and g represents Polarimetric SAR Image
The degree of coupling between symmetric part, b represent the irregularity factor of Polarimetric SAR Image, and e represents the surface distortion of Polarimetric SAR Image
Property, f represents the helicity of Polarimetric SAR Image, the asymmetric factor of l expression Polarimetric SAR Images；
(1b) obtains 12 primitive character parameters, respectively symmetrical factor a from coherence matrix T, irregularity factor b,
Configuration factor c, local curvature d, surface deflections sex factor e, spiral sex factor f, degree of coupling factor g, directional factors h are non-right
Claim factor l and represent Complex eigenvalues | c-id |, | h+ig |, | e+if |；
Each group profile of 12 groups of primitive characters of said extracted is shown as piece image by (1c), on every piece image
The sliding window of a 5*5 is selected, the pixel value of all the points represents the neighborhood of central point pixel in addition to central point pixel with window
The neighborhood characteristics of the same coordinate points of 12 width images are combined, form the neighborhood characteristics of each sample by feature.
Step 2, processing of taking the logarithm to primitive character and neighborhood characteristics, and be normalized.
(2a) takes the logarithm to primitive character and neighborhood characteristics processing, makes its noise Gaussian distributed:
It is shown according to documents and materials, it is mutually only with coherent speckle noise when Polarimetric SAR Image is interfered by coherent speckle noise
It is vertical, for this coherent speckle noise, gamma Gamma distributions are obeyed, transformation of being fetched to it when independently regarding number more than 3, this phase
Dry spot noise approximation Gaussian distributed.Therefore transformation of taking the logarithm to the primitive character and neighborhood characteristics of extraction, original spy can be made
The multiplicative noise of sign and neighborhood characteristics becomes additive noise, and approximate Gaussian distributed；
(2b) normalizes the primitive character after transformation and neighborhood characteristics, obtains the initial defeated of denoising autocoding DA networks
Enter data.
Step 3, the number of plies for determining denoising autocoding DA networks, each node layer number and data noise.
It is current research using denoising autocoding DA networks in the present invention and applies very extensive deep learning structure,
Deep learning and traditional shallow-layer learn it is maximum difference lies in：The depth of network structure is highlighted, the network number of plies generally reaches 4
On layer；The importance of feature learning is highlighted, by the nonlinear change of multiple hidden layers, DA networks possess than shallow-layer network
More excellent feature representation ability, and DA networks add the plant noise as initial data, can use and be destroyed
Input data reconstruct original data so that feature more robustness.DA networks are the autocoding layer groups by multilayer
Into autocoding is two layers of neural network, and interlayer connects entirely, connectionless in layer, the number of plies, each node layer number and data noise
It is determining as follows：
(3a) considers the classification performance of Polarimetric SAR Image, constructs the 4 layer depth networks based on denoising autocoding,
Including an input layer, two hidden layers and a classification layer, and layer of classifying obtains final denoising certainly using neural network NN
Dynamic encoding D A networks output, feature after the normalization extracted according to step 2, this example set input layer number as 300, the second layer
Node in hidden layer with third layer is 100, and classification layer is the neural network NN models for Classification of Polarimetric SAR Image,
Number of nodes is 9, and each node layer number for obtaining denoising autocoding DA networks is followed successively by 300,100,100,9；
(3b) obeys the characteristic of gamma distribution according to the coherent speckle noise of polarization SAR, is more than 3 in the case that independently regarding number,
It takes the logarithm gamma distribution as data noise.
Step 4, training denoising autocoding DA networks.
Training for depth network is always the difficult point of neural network research field, traditional based on gradient descent method
Global training method is generally difficult to prove effective for depth network, the problems such as local optimum is complete at present caused by training deep layer network
Office's very formidable obstacle of training method.
A kind of current thinking of effective trained depth network is to carry out order training method to depth network, then global micro-adjustment
The structural parameters of a network, detailed step are as follows：
(4a) to denoising autocoding DA networks input layer add in mean value be 0, the Gaussian noise that variance is 0.01；
(4b), to two layers of progress parameter adjustment before denoising autocoding DA networks, is obtained using reconstructed error minimum method
Weights and biasing between first layer and the second layer；
Second layer result after adjustment is input to third layer by (4c), reuse reconstructed error minimum method to denoising from
The second layer and third layer of dynamic encoding D A networks carry out parameter adjustment, obtain the weights between the second layer and third layer and biasing；
Third layer result after adjustment is input to the classification layer of denoising autocoding DA networks by (4d), with neural network NN
Parameter adjustment is carried out, obtains output result；
(4e) using back-propagation algorithm to the carry out integrated regulation of entire denoising autocoding DA networks, optimization denoising is certainly
The structural parameters of dynamic encoding D A networks, obtain trained denoising autocoding DA networks.
Step 5, using trained denoising autocoding DA networks, classify to Polarimetric SAR Image to be sorted, i.e.,
The original input data that step 2 obtains is input in the denoising autocoding DA networks that step 4 training obtains, and is exported to this polarization
The classification results of SAR image.
The effect of the present invention can be further illustrated by emulation experiment：
1. experiment condition and method
Experiment simulation environment：MATLAB 2013b, Windows7 Professional
Experimental method：Respectively existing depth belief network DBN, autocoding network and the method for the present invention, wherein preceding two
Kind method is classical deep learning method, and applied well in the classification of Polarimetric SAR Image.
Polarimetric SAR Image used in the emulation experiment of the present invention, size are 380*420 pixels, and image comes from NASA/
The L-band Holland Flevoland complete polarizations 4 that the AIRSAR in JPL laboratories is obtained regard data, and the resolution ratio of data is 12.1m*
The tag along sort figure of 6.7m, wherein Fig. 2 image thus.
2. experiment content and interpretation of result
Emulation one：Using existing depth belief network DBN to Classification of Polarimetric SAR Image, obtained classification results such as Fig. 3
It is shown；
Emulation two：Using existing autocoding network to Classification of Polarimetric SAR Image, obtained classification results such as Fig. 4 institutes
Show；
Emulation three：Using the method for the present invention to Classification of Polarimetric SAR Image, obtained classification results are as shown in Figure 5.
Interpretation of result：For Fig. 3 compared with Fig. 2, classifying quality is especially bad, and image boundary classification is fuzzy, many regions all by
To the influence of noise, region division is not fine, and accuracy is 92.1% ± 0.003, relatively low；Fig. 4 and Fig. 2 phases
Than overall effect is fine, but larger by influence of noise, and the top half classification of image is poor, and classification accuracy rate is
94.3% ± 0.04；Fig. 5 is compared with Fig. 2, and overall effect is very good, and noise exists on a small quantity, and smaller on result influence, on image
The effect of half part classification is fine, and lower half portion is slightly bad, and classification accuracy rate is 96.1% ± 0.04, and accuracy is higher.
Claims (2)
1. a kind of Classification of Polarimetric SAR Image method based on denoising autocoding DA networks, includes the following steps：
(1) one Polarimetric SAR Image to be sorted of input option extracts the primitive character and its neighborhood of the Polarimetric SAR Image
Feature：
(1.1) coherence matrix of Polarimetric SAR Image is decomposed according to the following formula：
Wherein T represents the coherence matrix of Polarimetric SAR Image, and i represents imaginary part unit,Polarimetric SAR Image it is each
The value of point is all the coherence matrix of a 3*3, and a represents the symmetrical factor of Polarimetric SAR Image, and c represents the configuration of Polarimetric SAR Image
The factor, d represent the local curvature of Polarimetric SAR Image, and h represents the directionality of Polarimetric SAR Image, and g represents that Polarimetric SAR Image is symmetrical
The degree of coupling between part, b represent the irregularity factor of Polarimetric SAR Image, and e represents the surface distortion of Polarimetric SAR Image, f
Represent the helicity of Polarimetric SAR Image, l represents the asymmetric factor of Polarimetric SAR Image；
(1.2) from coherence matrix T, 12 primitive character parameters, respectively symmetrical factor a, irregularity factor b, structure are obtained
Type factor c, local curvature d, surface deflections sex factor e, spiral sex factor f, degree of coupling factor g, directional factors h are asymmetric
Factor l and represent Complex eigenvalues | c-id |, | h+ig |, | e+if |；
(1.3) each group profile of 12 groups of primitive characters of said extracted is shown as piece image, is selected on every piece image
The sliding window of a 5*5 is selected, the pixel value of all the points represents that the neighborhood of central point pixel is special in addition to central point pixel with window
Sign, the neighborhood characteristics of the same coordinate points of 12 width images are combined, form the neighborhood characteristics of each sample；
(2) it takes the logarithm to primitive character and neighborhood characteristics processing, its noise is made to meet Gaussian Profile；
(3) number of plies, each node layer number and the data noise of denoising autocoding DA networks are determined：
(3.1) 4 layer networks based on denoising autocoding are established, this 4 layers are followed successively by input layer, two hidden layers and classification layer；
(3.2) the input layer number for specifying denoising autocoding is 300, and first node in hidden layer is 100, and second hidden
Number containing node layer is 100, and classification node layer number is 9；
(3.3) characteristic of gamma distribution is obeyed according to the coherent speckle noise of polarization SAR, is more than 3 in the case that independently regarding number, to gal
Horse distribution is taken the logarithm as data noise；
(4) training denoising autocoding DA networks：
(4a), with reference to figure, determines the classification number with reference to atural object in figure, 10% atural object mark is chosen in every class according to practical atural object
It is denoted as training sample, and the feature of training sample is input in the autocoding DA networks, carry out greediness instruction successively
Practice；
(4b) is finely tuned the structural parameters and noise parameter of denoising autocoding DA networks, is obtained using back-propagation algorithm BP
Trained denoising autocoding DA networks；
(5) trained denoising autocoding DA networks are utilized, classifies to Polarimetric SAR Image to be sorted, is polarized
The classification results of SAR image.
2. the Classification of Polarimetric SAR Image method according to claim 1 based on denoising autocoding DA networks, wherein described
Training denoising autocoding DA networks in step (4), carry out as follows：
(4.1) it is 0 to the input layer of denoising autocoding DA networks addition mean value, the Gaussian noise that variance is 0.01；
(4.2) the is obtained to two layers of progress parameter adjustment before denoising autocoding DA networks using reconstructed error minimum method
One layer of weights and biasing between the second layer；
(4.3) second layer result after adjustment is input to third layer, it is automatic to denoising reuses reconstructed error minimum method
The second layer and third layer of encoding D A networks carry out parameter adjustment, obtain the weights between the second layer and third layer and biasing；
(4.4) the third layer result after adjustment is input to the classification layer of denoising autocoding DA networks, with neural network NN into
Row parameter adjustment obtains output result；
(4.5) it is automatic that denoising is optimized to the carry out integrated regulation of entire denoising autocoding DA networks using back-propagation algorithm
The structural parameters of encoding D A networks obtain trained denoising autocoding DA networks.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510108639.6A CN104751172B (en) | 2015-03-12 | 2015-03-12 | The sorting technique of Polarimetric SAR Image based on denoising autocoding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510108639.6A CN104751172B (en) | 2015-03-12 | 2015-03-12 | The sorting technique of Polarimetric SAR Image based on denoising autocoding |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104751172A CN104751172A (en) | 2015-07-01 |
| CN104751172B true CN104751172B (en) | 2018-07-03 |
Family
ID=53590825
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510108639.6A Active CN104751172B (en) | 2015-03-12 | 2015-03-12 | The sorting technique of Polarimetric SAR Image based on denoising autocoding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104751172B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127230B (en) * | 2016-06-16 | 2019-10-01 | 上海海事大学 | Image-recognizing method based on human visual perception |
| CN106331433B (en) * | 2016-08-25 | 2020-04-24 | 上海交通大学 | Video denoising method based on deep recurrent neural network |
| CN109214401B (en) * | 2017-06-30 | 2020-10-16 | 清华大学 | SAR image classification method and device based on hierarchical automatic encoder |
| CN108388927B (en) * | 2018-03-26 | 2021-10-29 | 西安电子科技大学 | Small sample polarization SAR terrain classification method based on deep convolution twin network |
| CN108983804B (en) * | 2018-08-27 | 2020-05-22 | 燕山大学 | Biped robot gait planning method based on deep reinforcement learning |
| CN110335202A (en) * | 2019-04-08 | 2019-10-15 | 武汉理工大学 | A kind of underwater sonar image denoising method |
| CN111340719B (en) * | 2020-02-13 | 2023-03-31 | 华南农业大学 | Transient image data enhancement method based on full-connection automatic coding machine |
| CN111563423A (en) * | 2020-04-17 | 2020-08-21 | 西北工业大学 | Unmanned aerial vehicle image target detection method and system based on depth denoising automatic encoder |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2157544A1 (en) * | 2008-08-01 | 2010-02-24 | Julius-Maximilians-Universität Würzburg | System for adaptive removal of speckle noise in digital images and generation of a colour composite product based on automated analysis of speckle characteristics |
| CN103886336A (en) * | 2014-04-09 | 2014-06-25 | 西安电子科技大学 | Polarized SAR image classifying method based on sparse automatic encoder |
| CN104077599A (en) * | 2014-07-04 | 2014-10-01 | 西安电子科技大学 | Polarization SAR image classification method based on deep neural network |
-
2015
- 2015-03-12 CN CN201510108639.6A patent/CN104751172B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2157544A1 (en) * | 2008-08-01 | 2010-02-24 | Julius-Maximilians-Universität Würzburg | System for adaptive removal of speckle noise in digital images and generation of a colour composite product based on automated analysis of speckle characteristics |
| CN103886336A (en) * | 2014-04-09 | 2014-06-25 | 西安电子科技大学 | Polarized SAR image classifying method based on sparse automatic encoder |
| CN104077599A (en) * | 2014-07-04 | 2014-10-01 | 西安电子科技大学 | Polarization SAR image classification method based on deep neural network |
Non-Patent Citations (2)
| Title |
|---|
| CLASSIFICATION OF LAND COVER BASED ON DEEP BELIEF NETWORKS USING POLARIMETRIC RADARSAT-2 DATA;Qi Lv 等;《IGARSS 2014》;20141130;全文 * |
| SAR Automatic Target Recognition Based on Classifiers Fusion;Xin Yu 等;《2011 International Workshop on Multi-Platform/Multi-Sensor Remote Sensing and Mapping》;20110131;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104751172A (en) | 2015-07-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104751172B (en) | The sorting technique of Polarimetric SAR Image based on denoising autocoding | |
| Liu et al. | Deep depthwise separable convolutional network for change detection in optical aerial images | |
| Tang et al. | Compressed-domain ship detection on spaceborne optical image using deep neural network and extreme learning machine | |
| CN108052966B (en) | Remote sensing image scene automatic extraction and classification method based on convolutional neural network | |
| CN105184309A (en) | Polarization SAR image classification based on CNN and SVM | |
| CN108734171A (en) | A kind of SAR remote sensing image ocean floating raft recognition methods of depth collaboration sparse coding network | |
| Asokan et al. | Machine learning based image processing techniques for satellite image analysis-a survey | |
| Wu et al. | Multiscale CNN with autoencoder regularization joint contextual attention network for SAR image classification | |
| CN111080678B (en) | Multi-temporal SAR image change detection method based on deep learning | |
| CN112950780B (en) | Intelligent network map generation method and system based on remote sensing image | |
| Sun et al. | SAR image classification using greedy hierarchical learning with unsupervised stacked CAEs | |
| Shen et al. | Learning high-level concepts by training a deep network on eye fixations | |
| CN111274905A (en) | AlexNet and SVM combined satellite remote sensing image land use change detection method | |
| Diakite et al. | Hyperspectral image classification using 3D 2D CNN | |
| CN111563577B (en) | Unet-based intrinsic image decomposition method for skip layer frequency division and multi-scale identification | |
| Al-Amaren et al. | RHN: A residual holistic neural network for edge detection | |
| Feng et al. | Land-cover classification of high-resolution remote sensing image based on multi-classifier fusion and the improved Dempster–Shafer evidence theory | |
| Wang et al. | Facial expression recognition based on CNN | |
| Zhao et al. | Visible-infrared person re-identification based on frequency-domain simulated multispectral modality for dual-mode cameras | |
| Shi et al. | F 3 Net: Fast Fourier filter network for hyperspectral image classification | |
| Zeng et al. | Masanet: Multi-angle self-attention network for semantic segmentation of remote sensing images | |
| Srinitya et al. | Automated SAR image segmentation and classification using modified deep learning | |
| Hussein et al. | Semantic segmentation of aerial images using u-net architecture | |
| Li et al. | Spatial fuzzy clustering and deep auto-encoder for unsupervised change detection in synthetic aperture radar images | |
| Zhang et al. | Single-sample face recognition under varying lighting conditions based on logarithmic total variation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |