CN111080566A - Visible light and infrared image fusion method based on structural group double-sparse learning - Google Patents
Visible light and infrared image fusion method based on structural group double-sparse learning Download PDFInfo
- Publication number
- CN111080566A CN111080566A CN201911270444.6A CN201911270444A CN111080566A CN 111080566 A CN111080566 A CN 111080566A CN 201911270444 A CN201911270444 A CN 201911270444A CN 111080566 A CN111080566 A CN 111080566A
- Authority
- CN
- China
- Prior art keywords
- image
- sparse
- dictionary
- visible light
- learning
- 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
- 238000007500 overflow downdraw method Methods 0.000 title claims abstract description 14
- 230000004927 fusion Effects 0.000 claims abstract description 35
- 238000000034 method Methods 0.000 claims abstract description 22
- 239000011159 matrix material Substances 0.000 claims abstract description 21
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims abstract description 6
- 239000013598 vector Substances 0.000 claims description 13
- 230000000903 blocking effect Effects 0.000 claims description 2
- 238000010276 construction Methods 0.000 claims description 2
- 238000001514 detection method Methods 0.000 abstract description 2
- 238000003745 diagnosis Methods 0.000 abstract description 2
- 238000012544 monitoring process Methods 0.000 abstract description 2
- 230000008901 benefit Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 4
- 230000009466 transformation Effects 0.000 description 3
- 238000003331 infrared imaging Methods 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a visible light and infrared image fusion method based on structural group double-sparse learning. The method comprises the following steps: (1) performing sliding window processing on the input visible light and infrared images, searching similar blocks of an original image block, performing group vectorization, and establishing an image similar structure group matrix; (2) taking the image similar structure group matrix as a training sample, forming a base dictionary by using a Kronecker product of shear wavelets, obtaining a sparse dictionary through online learning, and performing linear reconstruction on the base dictionary and the sparse dictionary to obtain a final double-sparse dictionary; (3) and combining the double sparse dictionaries, performing group sparse solution on the image similar structure group by using SOMP to obtain a group sparse coefficient, and obtaining a final fusion image by using a large fusion rule and image reconstruction. The method solves the problem that the image fusion quality is low due to the fact that the existing sparse fusion algorithm ignores the correlation among image blocks and the dictionary adaptability is poor, and can be applied to the fields of remote sensing detection, medical diagnosis, intelligent driving, safety monitoring and the like.
Description
Technical Field
The invention relates to an image fusion method in the field of image processing, in particular to a visible light and infrared image fusion method based on structural group double sparse learning.
Background
Visible light and infrared imaging technologies have important applications in the aspects of remote sensing, medical diagnosis, intelligent driving, safety monitoring and the like. The visible light sensor can describe scene information of the environment through light reflection imaging, has higher spatial resolution, and is easily influenced by illumination conditions and weather change factors; the infrared sensor can reflect the radiation characteristics of the target and the background through thermal radiation imaging, but structural characteristics and texture information of the target are lacked. The two types of imaging utilize different physical characteristics of the target to detect, have strong complementarity, only fuse the two types of images, just can synthesize different imaging advantages, reduce information loss, do benefit to target identification processing and personnel and observe, satisfy practical demand. Therefore, image fusion is an important prerequisite for improving the detection and recognition levels of visible light and infrared imaging.
The key technology for realizing the fusion of the visible light image and the infrared image is to integrate the significant characteristics between the two images into one image so as to exert the integrated advantages of the visible light image and the infrared image. The multi-scale transformation fusion method is to approximate the salient feature information of the image by using a given mathematical model, but because the salient feature types of the image are complicated and changeable, the multi-scale transformation cannot extract the salient features of all types of the image. In order to improve the multi-scale transformation fusion effect, a redundant dictionary is constructed through online learning from the perspective of image signal sparsity, image signals are represented on the redundant dictionary through sparse representation, and the significant features of the images are described by using representation coefficients and atoms corresponding to the redundant dictionary. At present, the traditional sparse representation fusion model has two problems: firstly, in the process of dictionary learning and sparse coding, each image block is considered independently, and the correlation among the blocks is ignored, so that the sparse coding coefficient is inaccurate; secondly, the respective advantages of the analysis dictionary and the learning dictionary are not combined, and the adaptability of the dictionary is not strong.
Research shows that non-local similarity is an important characteristic of an image, and illustrates that many similar structures (such as detail and texture information) exist at different positions in the image, and the information contained in the similar structures is applied to image processing, so that the effect of the image processing can be improved, and the method is used in the fields of image denoising, compressed sensing, super-resolution and the like. In fact, visible light images have rich repetitive structures and a large amount of redundant information, and image blocks reflect local geometric structures and repeatedly appear at different positions of the images, and have non-local structural similarity. In the infrared image, most of the infrared image is a background area, the gray scale change of the infrared image is slow, and small background image blocks have strong correlation and obvious non-local structural similarity. Therefore, through the non-local similarity of the images, the structural similarity groups of the visible images and the infrared images are established, the correlation among the image blocks is further established, and the accuracy of sparse coding is improved.
At present, sparse representation dictionaries can be divided into two categories, parsing dictionaries and learning dictionaries. The analytic dictionary establishes a formulaic mathematical model for the data, so the analytic dictionary is highly structured, can realize fast numerical value, but has poor adaptability; the learning dictionary is used for learning through training samples, and is more adaptive, but the learning model is complex. Researches show that organic combination of an analysis dictionary and a training dictionary is realized, and the advantages of the two dictionaries are combined, so that the adaptability of the dictionary is improved, and the complexity of a model is reduced, which is an urgent need for sparse representation fusion development.
In summary, there is an urgent need for an image fusion method that can effectively establish the correlation of image blocks, improve the accuracy of sparse coding, enhance the applicability of redundant dictionaries, reduce the complexity of dictionary learning models, and further effectively improve the fusion effect of visible light and infrared images.
Disclosure of Invention
The invention provides a visible light and infrared image fusion method based on structural group double-sparse learning, and aims to solve the problem that the existing sparse representation fusion algorithm ignores the correlation among blocks and is poor in dictionary learning adaptability, so that the fusion quality of visible light and infrared images is poor.
The invention is realized by adopting the following technical scheme: a visible light and infrared image fusion method based on structural group double sparse learning comprises the following steps:
s1: performing sliding window processing on the input visible light and infrared images, searching for similar image blocks of the original image blocks, performing group vectorization on the original image blocks and the similar image blocks, and establishing an image similar structure group matrix;
s2: constructing a double-sparse learning model, forming a base dictionary by using a Kronecker product of a shear wavelet, obtaining a sparse dictionary through online learning, and performing linear reconstruction on the base dictionary and the sparse dictionary to obtain a final double-sparse dictionary;
s3: and combining the double sparse dictionaries, carrying out sparse solution on the image similar structure group by adopting the SOMP to obtain a group sparse coefficient, and obtaining a final fusion image by adopting a large fusion rule and image reconstruction.
The above-mentioned structure group matrix construction process of the visible light and infrared image is as follows: the method comprises the steps of carrying out image blocking on an input image through a sliding window, using Euclidean distance as a criterion, searching for similar image blocks of an original image block, enabling the original image block and the similar image blocks to form similar structure groups, arranging the image blocks in each structure similar group according to a column vector sequence, connecting the image block vectors end to end, and obtaining similar structure group matrixes of the visible light and the infrared image respectively.
The process for constructing the double sparse dictionaries facing the similar structure group matrix comprises the following steps: the method comprises the steps of taking a similar structure group matrix of visible light images and infrared images as training samples, obtaining a base dictionary by using a Kronecker product of shear wavelets, obtaining a learning dictionary by using a sparse learning model and adopting a sequential updating iterative learning method, and finally, carrying out linear reconstruction on the base dictionary and the learning dictionary to obtain a final double-sparse dictionary.
Compared with the existing sparse representation fusion technology, the method has the following advantages:
1. according to the method, the non-local similarity of the images is utilized, the image similar structure group matrix is constructed, the relation between the image blocks is established, the capability of capturing the remarkable features of the images by the dictionary atoms can be effectively enhanced, and the accuracy of dictionary learning sparse coding is improved.
2. The method comprises the steps of forming a base dictionary by a Kronecker product of shear wavelets, obtaining a sparse dictionary through online learning, and performing linear reconstruction on the base dictionary and the sparse dictionary to obtain a double-sparse dictionary; the double sparse dictionaries are combined with respective advantages of the analytic dictionary and the learning dictionary, complexity of a dictionary learning model is reduced, and applicability of the enhanced redundant dictionary is enhanced.
3. The image fusion method for image structure group matrix double sparse learning is established, the fusion effect is obvious, the method can also be applied to the fusion of multi-mode images, multi-focus images and medical images, and the method has high application value in the field of image fusion.
Drawings
FIG. 1 is a schematic diagram of the structure of the method of the present invention.
Fig. 2 shows a first set of visible light and infrared image fusion experiments, which sequentially include a visible light image, an infrared image and a fusion image from left to right.
Fig. 3 shows a second set of visible light and infrared image fusion experiments, which sequentially include a visible light image, an infrared image and a fusion image from left to right.
Fig. 4 shows a third set of visible light and infrared image fusion experiment, which sequentially includes a visible light image, an infrared image and a fusion image from left to right.
Fig. 5 shows a fourth set of the visible light and infrared image fusion experiment, which sequentially includes a visible light image, an infrared image and a fusion image from left to right.
Detailed Description
A visible light and infrared image fusion method based on structural group double sparse learning comprises the following steps:
s1: performing sliding window processing on the input visible light and infrared images, searching for similar image blocks of the original image blocks, performing group vectorization on the original image blocks and the similar image blocks, and establishing an image similar structure group matrix;
s11: the method comprises the steps of adopting a sliding window technology, wherein the size of a sliding window is N multiplied by N, the sliding step length is 1 pixel, and dividing a visible light image V and an infrared image I with the size of M multiplied by N into (M-N +1) · (N-N +1) image blocks.
S12: for each original image block piIn L × L neighborhood, the Euclidean distance is used as the measurement criterion to calculate the sum piS most similar image blocks, and the original image block piForming a similarity group g with s similar image blocksiThere are s +1 image blocks in each similarity group. For visible and infrared images, (M-N +1) · (N-N +1) groups of similar structures, respectively, are obtained, respectivelyAnd
s13: similar structure group for visible light and infrared imageAndfirstly, arranging image blocks according to the sequence of column vectors to obtain image block vectorsThen the s +1 image block vectors in the similar structure group are connected end to obtain a similar structure group vector with higher dimensionality
S14: combining each similar structure group vector as a matrix column to respectively obtain a similar structure group matrix of the visible light and the infrared image
S2: constructing a double-sparse learning model, forming a base dictionary by using a Kronecker product of a shear wavelet, obtaining a sparse dictionary through online learning, and performing linear reconstruction on the base dictionary and the sparse dictionary to obtain a final double-sparse dictionary;
s21: for two-dimensional separable shear wavelets, the Kronecker product of the shear wavelets is used to form a base dictionaryLet A be an element of Rw×mFor sparse learning dictionaries, X ∈ Rm×(M-n+1)·(N-n+1)For sparse coefficient matrix, similar structure of visible light and infrared image is grouped into matrixAs a training sample, the dual sparse online learning model can then be represented as:
wherein x isiIs an arbitrary row of the sparse coefficient matrix X, ajIs an arbitrary column of the sparse dictionary A, | | | | | non-woven phosphor0Is L0Norm, solving the number of nonzero elements in the vector (namely sparsity); and p and k respectively control the sparsity of the sparse coefficient X and the sparse dictionary A.
S22: by using a pair dictionary atom ajThe learning method of sequential update obtains the sparse learning dictionary a through iteration, and the learning process can be expressed as:
S23: and multiplying the base dictionary phi and the sparse learning dictionary A, and performing linear reconstruction to obtain a final double-sparse dictionary D.
S3: combining double sparse dictionaries, carrying out sparse solution on the image similar structure group matrix by using SOMP to obtain a group sparse coefficient, and obtaining a final fusion image by using a large fusion rule and image reconstruction;
s31: similar structure group matrix of visible light and infrared images by using SOMP algorithm in combination with double sparse dictionaries DPerforming sparse decomposition to respectively obtainGroup sparsity coefficients for visible and infrared imagesAnd
s32: the group sparse coefficient combination adopts a fusion rule with the maximum absolute value, and can be expressed as follows:
wherein the content of the first and second substances,is composed ofT is 1,2, …, m. Final fused set of similar structures vector passThus obtaining the product.
S33: for each obtained fusion similarity structure group vectorIt is equally divided into s +1 subvectors. And reconstructing each sub vector into an image block with the size of n multiplied by n, placing the image block at a position corresponding to the reconstructed fusion image, and averaging the value at each position of the reconstructed fusion image according to the pixel superposition times to obtain the final fusion image.
Claims (3)
1. A visible light and infrared image fusion method based on structural group double sparse learning is characterized by comprising the following steps:
s1: performing sliding window processing on the input visible light and infrared images, searching for similar image blocks of the original image blocks, performing group vectorization on the original image blocks and the similar image blocks, and establishing an image similar structure group matrix;
s2: constructing a double-sparse learning model, forming a base dictionary by using a Kronecker product of a shear wavelet, obtaining a sparse dictionary through online learning, and performing linear reconstruction on the base dictionary and the sparse dictionary to obtain a final double-sparse dictionary;
s3: and combining the double sparse dictionaries, carrying out sparse solution on the image similar structure group by adopting the SOMP to obtain a group sparse coefficient, and obtaining a final fusion image by adopting a large fusion rule and image reconstruction.
2. The visible light and infrared image fusion method based on structure group double sparse learning of claim 1, characterized in that the similar structure group matrix construction process of the visible light and infrared image is as follows: and carrying out image blocking on the input image through a sliding window, searching for similar image blocks of the original image block by adopting Euclidean distance as a criterion, forming similar structure groups by the original image block and the similar image blocks, arranging the image blocks in each structure similar group according to a column vector sequence, and connecting the image block vectors end to obtain a similar structure group matrix of the visible light and the infrared image.
3. The visible light and infrared image fusion method based on structure group double sparse learning according to claim 1 or 2, characterized in that the process of constructing the double sparse dictionary facing to the similar structure group matrix is as follows: the method comprises the steps of taking a similar structure group matrix of visible light images and infrared images as training samples, obtaining a base dictionary by using a Kronecker product of shear wavelets, obtaining a learning dictionary by using a sparse learning model and adopting a sequential updating iterative learning method, and finally, carrying out linear reconstruction on the base dictionary and the learning dictionary to obtain a final double-sparse dictionary.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911270444.6A CN111080566A (en) | 2019-12-12 | 2019-12-12 | Visible light and infrared image fusion method based on structural group double-sparse learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911270444.6A CN111080566A (en) | 2019-12-12 | 2019-12-12 | Visible light and infrared image fusion method based on structural group double-sparse learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111080566A true CN111080566A (en) | 2020-04-28 |
Family
ID=70313890
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911270444.6A Pending CN111080566A (en) | 2019-12-12 | 2019-12-12 | Visible light and infrared image fusion method based on structural group double-sparse learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111080566A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111652832A (en) * | 2020-07-09 | 2020-09-11 | 南昌航空大学 | Infrared and visible light image fusion method based on sliding window technology |
CN111815732A (en) * | 2020-07-24 | 2020-10-23 | 西北工业大学 | Method for coloring intermediate infrared image |
CN117218048A (en) * | 2023-11-07 | 2023-12-12 | 天津市测绘院有限公司 | Infrared and visible light image fusion method based on three-layer sparse smooth model |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104200451A (en) * | 2014-08-28 | 2014-12-10 | 西北工业大学 | Image fusion method based on non-local sparse K-SVD algorithm |
CN105761234A (en) * | 2016-01-28 | 2016-07-13 | 华南农业大学 | Structure sparse representation-based remote sensing image fusion method |
CN106251320A (en) * | 2016-08-15 | 2016-12-21 | 西北大学 | Remote sensing image fusion method based on joint sparse Yu structure dictionary |
CN110097501A (en) * | 2019-04-12 | 2019-08-06 | 武汉大学 | A kind of NDVI image interfusion method based on the sparse regularization of non-local mean gradient |
-
2019
- 2019-12-12 CN CN201911270444.6A patent/CN111080566A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104200451A (en) * | 2014-08-28 | 2014-12-10 | 西北工业大学 | Image fusion method based on non-local sparse K-SVD algorithm |
CN105761234A (en) * | 2016-01-28 | 2016-07-13 | 华南农业大学 | Structure sparse representation-based remote sensing image fusion method |
CN106251320A (en) * | 2016-08-15 | 2016-12-21 | 西北大学 | Remote sensing image fusion method based on joint sparse Yu structure dictionary |
CN110097501A (en) * | 2019-04-12 | 2019-08-06 | 武汉大学 | A kind of NDVI image interfusion method based on the sparse regularization of non-local mean gradient |
Non-Patent Citations (3)
Title |
---|
RUBINSTEIN R等: ""Double Sparsity:Learning Sparse Dictionaries for Sparse Signal Approximation"", 《IEEE TRANSAXTIONS ON SIGNAL PROCESSING》 * |
张晓等: ""基于结构组稀疏表示的遥感图像融合"", 《中国图象图形学报》 * |
高珊等: ""基于双稀疏字典的医学图像融合算法及在脑血管疾病诊断中的应用"", 《北京生物医学工程》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111652832A (en) * | 2020-07-09 | 2020-09-11 | 南昌航空大学 | Infrared and visible light image fusion method based on sliding window technology |
CN111652832B (en) * | 2020-07-09 | 2023-05-12 | 南昌航空大学 | Infrared and visible light image fusion method based on sliding window technology |
CN111815732A (en) * | 2020-07-24 | 2020-10-23 | 西北工业大学 | Method for coloring intermediate infrared image |
CN111815732B (en) * | 2020-07-24 | 2022-04-01 | 西北工业大学 | Method for coloring intermediate infrared image |
CN117218048A (en) * | 2023-11-07 | 2023-12-12 | 天津市测绘院有限公司 | Infrared and visible light image fusion method based on three-layer sparse smooth model |
CN117218048B (en) * | 2023-11-07 | 2024-03-08 | 天津市测绘院有限公司 | Infrared and visible light image fusion method based on three-layer sparse smooth model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109376804B (en) | Hyperspectral remote sensing image classification method based on attention mechanism and convolutional neural network | |
Zhang et al. | Sparse representation based multi-sensor image fusion for multi-focus and multi-modality images: A review | |
CN109741256B (en) | Image super-resolution reconstruction method based on sparse representation and deep learning | |
Shekhar et al. | Analysis sparse coding models for image-based classification | |
Ji et al. | Nonlocal tensor completion for multitemporal remotely sensed images’ inpainting | |
CN111080566A (en) | Visible light and infrared image fusion method based on structural group double-sparse learning | |
Yang et al. | Multitask dictionary learning and sparse representation based single-image super-resolution reconstruction | |
CN106485688B (en) | High spectrum image reconstructing method neural network based | |
Marivani et al. | Multimodal deep unfolding for guided image super-resolution | |
CN107301630B (en) | CS-MRI image reconstruction method based on ordering structure group non-convex constraint | |
WO2017110836A1 (en) | Method and system for fusing sensed measurements | |
Zou et al. | Robust compressive sensing of multichannel EEG signals in the presence of impulsive noise | |
Aghamaleki et al. | Image fusion using dual tree discrete wavelet transform and weights optimization | |
CN108257093A (en) | The single-frame images ultra-resolution method returned based on controllable core and Gaussian process | |
CN114693577B (en) | Infrared polarized image fusion method based on Transformer | |
CN109447933B (en) | The infrared and visible light image fusion method decomposed based on peculiar synchronizing information | |
Wu et al. | A distributed fusion framework of multispectral and panchromatic images based on residual network | |
Heiser et al. | Compressive hyperspectral image reconstruction with deep neural networks | |
CN108596866B (en) | Medical image fusion method based on combination of sparse low-rank decomposition and visual saliency | |
Liu et al. | GJTD-LR: A trainable grouped joint tensor dictionary with low-rank prior for single hyperspectral image super-resolution | |
Ye et al. | Cross-scene hyperspectral image classification based on DWT and manifold-constrained subspace learning | |
CN116071226B (en) | Electronic microscope image registration system and method based on attention network | |
Niresi et al. | Robust hyperspectral inpainting via low-rank regularized untrained convolutional neural network | |
Shabanzade et al. | MR and PET image fusion using nonparametric Bayesian joint dictionary learning | |
Zeng et al. | U-net-based multispectral image generation from an rgb image |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200428 |