CN106600553B - DEM super-resolution method based on convolutional neural network - Google Patents
DEM super-resolution method based on convolutional neural network Download PDFInfo
- Publication number
- CN106600553B CN106600553B CN201611159517.0A CN201611159517A CN106600553B CN 106600553 B CN106600553 B CN 106600553B CN 201611159517 A CN201611159517 A CN 201611159517A CN 106600553 B CN106600553 B CN 106600553B
- Authority
- CN
- China
- Prior art keywords
- resolution
- dem
- data
- neural network
- super
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 76
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000000605 extraction Methods 0.000 claims abstract description 15
- 238000013528 artificial neural network Methods 0.000 claims abstract description 6
- 238000005070 sampling Methods 0.000 claims description 27
- 238000012545 processing Methods 0.000 claims description 19
- 230000005284 excitation Effects 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 7
- 238000011478 gradient descent method Methods 0.000 claims description 5
- 230000015556 catabolic process Effects 0.000 claims description 4
- 238000006731 degradation reaction Methods 0.000 claims description 4
- 239000007787 solid Substances 0.000 claims description 2
- 230000006870 function Effects 0.000 description 9
- 238000013507 mapping Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
Classifications
-
- G06T5/73—
-
- 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
- G06T2207/20084—Artificial neural networks [ANN]
Abstract
the invention discloses a DEM super-resolution method based on a convolutional neural network, which comprises the following steps of: (1) training a convolution neural network for super-resolution obtained according to low-resolution image data and high-resolution image data which correspond to each other in advance; (2) expanding the low-resolution DEM data to be processed by utilizing an interpolation method to obtain quasi-high-resolution DEM data with the same size as the expected high-resolution DEM data; (3) obtaining a gradient map of the quasi-high resolution DEM data by using an edge extraction operator; (4) inputting the gradient map into a convolution neural network for super-resolution to obtain an estimated gradient map of high-resolution DEM data; (5) and reconstructing a height map of the high-resolution DEM based on the constraints of the estimated gradient map and the low-resolution DEM data to be processed. The super-resolution method has strong robustness and high reconstruction result precision.
Description
Technical Field
the invention belongs to the technical field of topographic mapping, and particularly relates to a DEM super-resolution method based on a convolutional neural network.
Background
a Digital Elevation Model (DEM) is a branch of a Digital terrain Model, which is a Digital Model that represents the Elevation of the ground in the form of a set of ordered arrays of values. The method has wide application requirements in economy and national defense construction, for example, in flight path planning, people hope to utilize more accurate terrain data, and the planning result is optimal as far as possible, so that the pursuit of people for high-precision terrain models is a permanent theme. To obtain a highly accurate terrain model, two methods are generally used. One method generally uses higher precision measurement devices such as high resolution remote sensing images and achieves the goal of higher precision through dense measurements. The method has high production cost, and particularly for surveying and mapping the submarine topography, the cost is increased remarkably compared with the method for surveying and mapping the ground surface by using the satellite remote sensing image. The second method adopts a super-resolution strategy, improves the accuracy and resolution of the DEM and reduces the cost of acquiring high-resolution data by processing low-resolution data, thereby attracting the attention of a large number of researchers.
according to the analysis, the existing super-resolution strategy mainly adopts a non-local learning method, the method adopts a traditional super-resolution method based on learning, according to the manifold learning principle, the similarity between the test DEM data and the low-resolution data in the sample library is calculated, and the super-resolution processing is carried out according to the similarity and the high-resolution data corresponding to the similar DEM. This method improves the accuracy of high resolution DEM data to some extent, but has the disadvantage that it is relatively less adaptive and relies heavily on similarity measurements. In addition, to establish a convolutional neural network-based DEM super-resolution method, a large number of DEM samples are required, and a high-precision DEM is difficult to acquire relative to an image sample because most of the high-resolution DEM is confidential. Therefore, the establishment of the DEM super-resolution method which does not depend on high-precision DEM data and has good robustness is of great significance.
Disclosure of Invention
aiming at the defects or improvement requirements of the prior art, the invention aims to provide a DEM super-resolution method based on a convolutional neural network, wherein the key low-resolution DEM data processing process, the training learning mode of the convolutional neural network and the like are improved, so that the super-resolution method has strong robustness and the reconstruction result has high precision; in addition, the method meets the precision of the convolutional neural network trained by the image sample by virtue of the relatively small gradient dynamic range of the DEM, thereby avoiding obtaining high-precision DEM sample data and greatly reducing the production cost.
In order to achieve the above object, according to the present invention, there is provided a DEM super resolution method based on a convolutional neural network, comprising the steps of:
(1) training a convolution neural network for super-resolution obtained according to low-resolution image data and high-resolution image data which correspond to each other in advance;
(2) expanding the low-resolution DEM data to be processed by s times by utilizing an interpolation method to obtain quasi-high-resolution DEM data with the same size as the expected high-resolution DEM data;
(3) obtaining a gradient map of the quasi-high resolution DEM data by utilizing an edge extraction operator;
(4) Inputting the gradient map obtained in the step (3) into the convolutional neural network for super-resolution obtained in the step (1) to obtain an estimated gradient map of high-resolution DEM data and a corresponding estimated gradient value;
(5) And (4) reconstructing a height map of the high-resolution DEM based on the estimated gradient map obtained in the step (4) and the constraint of the low-resolution DEM data to be processed.
As a further preferred aspect of the present invention, the step (1) specifically includes the steps of:
(1-1) performing degradation processing on high-resolution image data to obtain corresponding low-resolution image data, then extracting the high-resolution image data and a gradient map of the low-resolution image data by using an edge extraction operator, then dividing each gradient map into a plurality of gradient blocks, and then selecting the corresponding gradient blocks of the high-resolution image data and the corresponding gradient blocks of the low-resolution image data as training samples;
(1-2) constructing a convolutional neural network and setting model parameters;
(1-3) training the convolutional neural network in the step (1-2) according to the training sample obtained in the step (1-1) to obtain a convolutional neural network for super-resolution.
as a further preferred aspect of the present invention, the convolutional neural network in the step (1-2) includes a plurality of convolutional layers, any two adjacent convolutional layers are connected through an excitation layer; wherein the first convolutional layer is used for inputting low-resolution image data, and the last convolutional layer is used for outputting high-resolution image data.
As a further preferred aspect of the present invention, the step (5) specifically includes the steps of:
(5-1) constructing a least square function for the target high resolution DEM data based on the estimated gradient map and the low resolution DEM data constraint to be processed;
(5-2) finding an optimal solution of the least square function in the step (5-1) by using an iteratively updated solution method;
(5-3) if the average reconstruction error between the data obtained after the optimal solution is subjected to the down-sampling processing and the low-resolution DEM data to be processed exceeds a preset threshold Th or the iteration frequency does not reach a preset requirement, returning to the step (5-2); otherwise, the optimal solution corresponds to the final high-resolution DEM data, and the optimal solution is used for reconstructing a height map of the high-resolution DEM.
As a further preferred aspect of the present invention, in the step (5-1), the least square function isWherein X is the low-resolution DEM data to be processed; y is the high resolution DEM data for the target, Y is the gradient corresponding to Y,estimating gradient values for the high resolution DEM data obtained in step (4); ↓sRepresents s times down-sampling processing, Y ↓sThe data is obtained by performing s-time downsampling processing on Y; β is a predetermined weighting factor.
As a further preferred aspect of the present invention, the solution method of the iterative update in the step (5-2) is a gradient descent method; preferably, Y after t +1 th iteration updatet+1satisfies the following conditions:
Wherein, YtIs the data after the t iteration update; τ is a preset iteration step; y is the Y after the t +1 th iteration updatet+1(ii) a X is the one to be treatedLow resolution DEM data; ↓sRepresents s times down-sampling processing, Y ↓sthe data is obtained by performing s-time downsampling processing on Y; ↓ (particulate solid) bearingsrepresents the s-fold upsampling process, (X-Y ↓s)↑sTo (X-Y ↓)s) Performing sampling processing on the data by s times; y is the gradient corresponding to Y,Estimating gradient values for the high resolution DEM data obtained in step (4); beta is a preset weight factor; div is the divergence operation.
In a further preferred embodiment of the present invention, in the step (5-3), Th is preferably 5;
in the step (5-1) and the step (5-2), β is preferably 0.03.
As a further preferred embodiment of the present invention, in the step (5-2), the maximum number of iterations is preferably 150, and the iteration step τ is preferably 0.2.
As a further preferred aspect of the present invention, the upsampling and the downsampling are performed by an interpolation method or an alternate sampling method; preferably, the interpolation method is nearest neighbor interpolation, bilinear interpolation or bicubic interpolation;
The edge extraction operator may be a Sobel operator, a Roberts operator, a Prewitt operator, or a Canny operator.
As a further preferred aspect of the present invention, in the step (3), the gradient map of the quasi-high resolution DEM data includes edge maps in an X direction and a Y direction, and the X direction and the Y direction are perpendicular to each other.
compared with the prior art, the technical scheme of the invention can reconstruct the high-resolution DEM data by training the convolution neural network for the super-resolution according to the low-resolution image data and the high-resolution image data which are mutually corresponding in advance (wherein the low-resolution image data can be obtained by carrying out downsampling on the high-resolution image data), and then processing the initial low-resolution DEM data (namely the low-resolution DEM data to be processed) by utilizing the convolution neural network for the super-resolution.
the DEM super-resolution method based on the convolutional neural network mainly comprises the following steps: firstly, obtaining a large amount of high-resolution data of an image, performing degradation operations such as down-sampling and blurring on all images in an image database to obtain corresponding low-resolution data, then respectively extracting gradient images of the high-resolution data and the low-resolution data by using an edge extraction operator, cutting the gradient images into small blocks, and taking the corresponding gradient blocks of the high-resolution data and the low-resolution data as training samples; then, based on the obtained training sample, establishing and training a convolutional neural network for super-resolution mapping from low resolution to high resolution; then, expanding the low-resolution DEM data by utilizing an interpolation method to obtain a quasi-high-resolution DEM with the same size as the expected high-resolution DEM data; then, obtaining a gradient map of the quasi-high resolution DEM data by using an edge extraction operator; then, inputting the gradient map into a trained convolutional neural network to obtain an estimated gradient map of the high-resolution DEM data; and finally, reconstructing a height map of the high-resolution DEM based on the estimated gradient map and the constraint of the original low-resolution DEM data. The invention can reconstruct the high-resolution DEM data by utilizing the image data, thereby solving the problem that the high-resolution DEM data is difficult to obtain in large quantity, and the reconstruction result is clear and has high accuracy.
Generally, compared with the prior art, the above technical solution conceived by the present invention has the following beneficial effects:
1. the learning method based on the convolutional neural network is introduced into DEM reconstruction, and once trained, the learning method can be used for super-resolution of all DEM data, so that the robustness of the algorithm is improved.
2. The gradient super-resolution of the DEM is realized by adopting the CNN (convolutional neural network) trained by the image sample, so that the DEM super-resolution is reconstructed, the working content of collecting high-precision DEM sample data is avoided, and the production cost is greatly reduced.
3. The thought based on the gradient is adopted to avoid the defect of larger result error caused by the fact that the dynamic range of the DEM is higher than that of the image, thereby improving the accuracy of the algorithm, leading the reconstruction result to be clear and leading the accuracy to be higher.
Drawings
FIG. 1 is a flow chart of a DEM super resolution method according to an embodiment of the invention;
Fig. 2 is a diagram of a convolutional neural network (i.e., CNN) for super resolution;
Fig. 3 is a diagram comparing the DEM super resolution method of the embodiment of the present invention with the conventional method.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
As shown in fig. 1, the DEM super-resolution method based on the convolutional neural network according to the embodiment of the present invention includes the following steps:
(1) the low resolution DEM data is expanded s times using an interpolation method to obtain quasi-high resolution DEM data of the same scale size (i.e., the same image size) as the desired high resolution DEM data.
the interpolation method is nearest neighbor interpolation, bilinear interpolation or bicubic interpolation.
(2) And obtaining a gradient map of the quasi-high-resolution DEM data by using an edge extraction operator, wherein the gradient map comprises edge maps in the X direction and the Y direction.
The edge extraction operator can be a Sobel operator, a Roberts operator, a Prewitt operator or a Canny operator.
(3) Inputting the gradient map into a convolutional neural network for super-resolution obtained by training according to high-resolution and low-resolution image data to obtain an estimated gradient map of the high-resolution DEM data;
The convolutional neural network for super-resolution obtained by training the high-resolution image data and the low-resolution image data is obtained according to the following steps:
(3-1) acquiring high-resolution data of a plurality of images (the larger the number is, the better the quality is), performing degradation operations such as down-sampling and up-sampling on all images in an image database to obtain corresponding low-resolution image data, extracting a gradient image from the high-resolution image data and the low-resolution image data by using an edge extraction operator, cutting the gradient image into small blocks, and taking the gradient blocks of the corresponding high-resolution image data and the corresponding low-resolution image data as training samples, wherein the up-sampling and the down-sampling can adopt a bicubic interpolation method or an alternate point sampling method;
(3-2) constructing a convolutional neural network and setting model parameters, and fig. 2 shows a model of the convolutional neural network, which includes a convolutional layer, an excitation layer, and a convolutional layer connected in sequence (wherein each excitation layer is not shown in fig. 2). And finally, outputting the convolution layer to obtain the corresponding high-resolution data. Then setting model parameters, wherein the learning rate is preferably set to be 0.001, and the learning rate is reduced to 1/5 of the original learning rate per 10000 times of training;
(3-3) training the convolutional neural network obtained in the step (3-2) according to the gradient block of the high-resolution and low-resolution image data obtained in the step (3-1) to obtain the convolutional neural network for super-resolution;
preferably, the convolutional neural network model in step (3-2) includes a convolutional layer, an excitation layer, a convolutional layer, an excitation layer and a convolutional layer connected in sequence. And finally, outputting the convolution layer to obtain the corresponding high-resolution data.
(4) And (4) reconstructing a height map of the high-resolution DEM through the constraints of the estimated gradient map obtained in the step (3) and the original low-resolution DEM data.
The specific steps of the height reconstruction process of the high-resolution DEM are preferably as follows:
(4-1) constructing a least square function based on the gradient map and low-resolution DEM data constraint;
(4-2) searching by using an iterative updating solution method to obtain an optimal solution;
(4-3) if the average reconstruction error of the downsampled version of the optimal solution and the original low resolution DEM data exceeds a certain threshold Th or the iteration time t is not enough, returning to the step (4-2).
Wherein said minimumThe multiplication function is preferablyX is the raw low resolution data, Y is the desired high resolution DEM,. Y is its corresponding gradient,and (4) for the estimated gradient obtained in the step (3), s is a sampling rate, ↓ represents downsampling, and beta is a weight and is used for controlling the balance coefficient of the original DEM weight and the gradient area. The strategy for iterative updating is preferably a gradient descent method. Such as:
Where t is the number of iterations, τ is the iteration step, ≈ up-sampling operation, and div is divergence operation.
The parameter threshold Th in the height reconstruction of the high-resolution DEM is preferably 5, and the weight β is 0.03. The number of iterations t is 150 and the iteration step τ is 0.2.
Preferably, the upsampling and downsampling may adopt an interpolation method or an alternate sampling method.
Preferably, the interpolation method is nearest neighbor interpolation, bilinear interpolation or bicubic interpolation.
Preferably, the edge extraction operator can be a Sobel operator, a Roberts operator, a Prewitt operator or a Canny operator.
the following are specific examples:
Example 1
This embodiment 1 includes the following steps:
(1) and expanding the low-resolution DEM data by s times by using a bicubic interpolation method to obtain quasi-high-resolution DEM data with the same size as the expected high-resolution DEM data.
(2) And extracting by using a Sobel operator to obtain edge maps in the X direction and the Y direction of the quasi-high-resolution DEM data.
(3) Inputting the gradient map into a convolutional neural network for super-resolution obtained by training according to high-resolution and low-resolution image data to obtain an estimated gradient map of the high-resolution DEM data;
The convolutional neural network for super-resolution obtained by training the high-resolution image data and the low-resolution image data is obtained according to the following steps:
(3-1) acquiring a large amount of high-resolution image data, carrying out first-time and second-time downsampling and then-time upsampling operations on all images in an image database to obtain corresponding low-resolution image data, extracting a gradient image from the high-resolution and low-resolution image data by using a Sobel operator, cutting the gradient image into small blocks, and taking the gradient blocks of the corresponding high-resolution and low-resolution image data as a training sample method, wherein the size of each low-resolution gradient block is 33, and the size of each high-resolution gradient block is 21;
(3-2) constructing a convolutional neural network, wherein a convolutional neural network model is shown in FIG. 2 and comprises a convolutional layer, an excitation layer, a convolutional layer, an excitation layer and a convolutional layer which are connected in sequence. And finally, outputting the convolution layer to obtain the corresponding high-resolution data.
Setting the model parameters as follows: the first convolutional layer template size is set to 1 × 9 × 9 × 64, the second convolutional layer template size is set to 64 × 1 × 1 × 32, and the last convolutional layer template size is set to 32 × 5 × 5 × 1. The excitation layers all adopt ReLU (max (0, x)).
(3-3) training the convolutional neural network obtained in the step (3-2) according to the gradient block of the high-resolution and low-resolution image data obtained in the step (3-1) to obtain the convolutional neural network for super-resolution;
the learning rate was set to 0.001, and the learning rate dropped to 1/5 of the original learning rate per 10000 times of training. And carrying out iterative updating on the weight by adopting a random gradient descent method.
(4) And (4) reconstructing a height map of the high-resolution DEM through the constraints of the estimated gradient map obtained in the step (3) and the original low-resolution DEM data.
The specific steps of the height reconstruction process of the high-resolution DEM are as follows:
(4-1) constructing a least square function based on the gradient map and low-resolution DEM data constraint;
(4-2) searching an optimal solution by using an iterative updating solution method;
(4-3) if the average reconstruction error of the downsampled version of the optimal solution and the original low resolution DEM data exceeds a certain threshold Th or the iteration time t is not enough, returning to the step (4-2).
Wherein the least square function is preferablyX is the raw low resolution data, Y is the desired high resolution DEM,. Y is its corresponding gradient,And (4) for the estimated gradient obtained in the step (3), s is a sampling rate, ↓ represents downsampling, and beta is a weight and is used for controlling the balance coefficient of the original DEM weight and the gradient area. The strategy for iterative updating is preferably a gradient descent method. Such as:
where t is the number of iterations, τ is the iteration step, ≈ up-sampling operation, and div is divergence operation.
and the parameter threshold Th in the height reconstruction of the high-resolution DEM is set to be 5, and the weight beta is 0.03. The number of iterations t is 150 and the iteration step τ is 0.2.
Table 1 shows the qualitative quality evaluation of the results of 2 sets of test data under different s-fold amplification and the results of the conventional bicubic interpolation method. Fig. 3 is a diagram showing the result of DEM2 at 4 times magnification. In which fig. 3(a) is a real graph, fig. 3(b) is a bicubic interpolation result, and fig. 3(c) is a test result.
TABLE 1
as is apparent from table 1 and fig. 3, the obtained result has a significant accuracy improvement and more obvious details than the conventional interpolation method under different data. Thus also verifying the rationality of our approach.
each function (such as various edge extraction operators, argmin (Y) functions) which are not described in detail in the invention can be a conventional definition in the mathematical field; the up-sampling and down-sampling data processing methods (such as an interpolation method and a dot-separating sampling method) can refer to the prior art in the field of image processing. For example,representation matrix (X-Y ↓)s) The second paradigm of (1).
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. a DEM super-resolution method based on a convolutional neural network is characterized by comprising the following steps:
(1) training a convolution neural network for super-resolution obtained according to low-resolution image data and high-resolution image data which correspond to each other in advance;
(2) expanding the low-resolution DEM data to be processed by s times by utilizing an interpolation method to obtain quasi-high-resolution DEM data with the same size as the expected high-resolution DEM data;
(3) Obtaining a gradient map of the quasi-high resolution DEM data by utilizing an edge extraction operator;
(4) inputting the gradient map obtained in the step (3) into the convolutional neural network for super-resolution obtained in the step (1) to obtain an estimated gradient map of high-resolution DEM data and a corresponding estimated gradient value;
(5) Reconstructing a height map of the high-resolution DEM based on the estimated gradient map obtained in the step (4) and the constraint of the low-resolution DEM data to be processed;
Wherein, the step (1) specifically comprises the following steps:
(1-1) performing degradation processing on high-resolution image data based on an image sample to obtain corresponding low-resolution image data, then extracting the high-resolution image data and a gradient map of the low-resolution image data by using an edge extraction operator, then dividing each gradient map into a plurality of gradient blocks, and then selecting the corresponding gradient blocks of the high-resolution image data and the corresponding gradient blocks of the low-resolution image data as training samples;
(1-2) constructing a convolutional neural network and setting model parameters;
(1-3) training the convolutional neural network in the step (1-2) according to the training sample obtained in the step (1-1) to obtain a convolutional neural network for super-resolution;
The step (5) specifically comprises the following steps:
(5-1) constructing a least square function for the target high resolution DEM data based on the estimated gradient map and the low resolution DEM data constraint to be processed;
(5-2) finding an optimal solution of the least square function in the step (5-1) by using an iteratively updated solution method;
(5-3) if the average reconstruction error between the data obtained after the optimal solution is subjected to the down-sampling processing and the low-resolution DEM data to be processed exceeds a preset threshold Th or the iteration frequency does not reach a preset requirement, returning to the step (5-2); otherwise, the optimal solution corresponds to the final high-resolution DEM data, and the optimal solution is used for reconstructing a height map of the high-resolution DEM.
2. The DEM super resolution method based on convolutional neural network as claimed in claim 1, wherein said convolutional neural network in step (1-2) comprises a plurality of convolutional layers, any two adjacent convolutional layers are connected by an excitation layer; wherein the first convolutional layer is used for inputting low-resolution image data, and the last convolutional layer is used for outputting high-resolution image data.
3. the convolutional neural network-based of claim 1The DEM super-resolution method is characterized in that in the step (5-1), the least square function isWherein X is the low-resolution DEM data to be processed; y is the high resolution DEM data of the target,the gradient corresponding to the Y is the gradient,estimating gradient values for the high resolution DEM data obtained in step (4); ↓sRepresents s times down-sampling processing, Y ↓sthe data is obtained by performing s-time downsampling processing on Y; β is a predetermined weighting factor.
4. The convolutional neural network based DEM super resolution method as claimed in claim 1, wherein the solution method of the iterative update in step (5-2) is a gradient descent method; y after t +1 th iteration updatet+1Satisfies the following conditions:
Wherein, YtIs the data after the t iteration update; τ is a preset iteration step; y is the Y after the t +1 th iteration updatet+1(ii) a X is the low-resolution DEM data to be processed; ↓sRepresents s times down-sampling processing, Y ↓sthe data is obtained by performing s-time downsampling processing on Y; ↓ (particulate solid) bearingsRepresents the s-fold upsampling process, (X-Y ↓s)↑sto (X-Y ↓)s) Performing sampling processing on the data by s times;the gradient corresponding to the Y is the gradient,estimating gradient values for the high resolution DEM data obtained in step (4); beta is a preset weight factor; div is the divergence operation.
5. The DEM super resolution method based on convolutional neural network as claimed in claim 3, wherein in said step (5-3), Th is 5;
in the step (5-1) and the step (5-2), β is 0.03.
6. the DEM super resolution method based on convolutional neural network as claimed in claim 4, wherein in said step (5-3), Th is 5;
In the step (5-1) and the step (5-2), β is 0.03.
7. the DEM super resolution method based on convolutional neural network as claimed in claim 4, wherein in step (5-2), the maximum number of iterations is 150, and the iteration step τ is 0.2.
8. the DEM super-resolution method based on the convolutional neural network as claimed in claim 4, 6 or 7, wherein the up-sampling process is an interpolation method or a point-separated sampling method; the interpolation method is nearest neighbor interpolation, bilinear interpolation or bicubic interpolation;
the edge extraction operator may be a Sobel operator, a Roberts operator, a Prewitt operator, or a Canny operator.
9. The DEM super-resolution method based on the convolutional neural network as claimed in any one of claims 1-7, wherein the down-sampling process is an interpolation method or an alternate sampling method; the interpolation method is nearest neighbor interpolation, bilinear interpolation or bicubic interpolation;
the edge extraction operator may be a Sobel operator, a Roberts operator, a Prewitt operator, or a Canny operator.
10. The convolutional neural network-based DEM super-resolution method as claimed in claim 1, wherein in step (3), the gradient map of the quasi-high resolution DEM data comprises edge maps in an X direction and a Y direction, and the X direction and the Y direction are perpendicular to each other.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611159517.0A CN106600553B (en) | 2016-12-15 | 2016-12-15 | DEM super-resolution method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611159517.0A CN106600553B (en) | 2016-12-15 | 2016-12-15 | DEM super-resolution method based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106600553A CN106600553A (en) | 2017-04-26 |
| CN106600553B true CN106600553B (en) | 2019-12-17 |
Family
ID=58801511
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611159517.0A Expired - Fee Related CN106600553B (en) | 2016-12-15 | 2016-12-15 | DEM super-resolution method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106600553B (en) |
Families Citing this family (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169927B (en) * | 2017-05-08 | 2020-03-24 | 京东方科技集团股份有限公司 | Image processing system, method and display device |
| CN107689034B (en) * | 2017-08-16 | 2020-12-01 | 清华-伯克利深圳学院筹备办公室 | Denoising method and denoising device |
| CN107578375B (en) * | 2017-08-21 | 2020-10-02 | 北京陌上花科技有限公司 | Image processing method and device |
| CN107945134B (en) * | 2017-11-30 | 2020-10-09 | 北京小米移动软件有限公司 | Image processing method and device |
| CN110321913B (en) * | 2018-03-30 | 2023-07-25 | 杭州海康威视数字技术股份有限公司 | Text recognition method and device |
| CN109584164B (en) * | 2018-12-18 | 2023-05-26 | 华中科技大学 | Medical image super-resolution three-dimensional reconstruction method based on two-dimensional image transfer learning |
| CN112419146B (en) * | 2019-08-20 | 2023-12-29 | 武汉Tcl集团工业研究院有限公司 | Image processing method and device and terminal equipment |
| CN110706166B (en) * | 2019-09-17 | 2022-03-18 | 中国科学院空天信息创新研究院 | Image super-resolution reconstruction method and device for sharpening label data |
| CN111008640B (en) * | 2019-10-17 | 2024-03-19 | 平安科技（深圳）有限公司 | Image recognition model training and image recognition method, device, terminal and medium |
| CN110827375B (en) * | 2019-10-31 | 2023-05-30 | 湖北大学 | Infrared image true color coloring method and system based on low-light-level image |
| CN111754405B (en) * | 2020-06-22 | 2023-08-08 | 北京大学深圳研究生院 | Image resolution reduction and restoration method, device and readable storage medium |
| CN112200751A (en) * | 2020-10-23 | 2021-01-08 | 华强方特（深圳）电影有限公司 | Image enhancement method |
| CN112669209B (en) * | 2020-12-24 | 2023-06-16 | 华中科技大学 | Three-dimensional medical image super-resolution reconstruction method and system |
| CN112907733A (en) * | 2021-02-23 | 2021-06-04 | 北京华清易通科技有限公司 | Method and device for reconstructing three-dimensional model and three-dimensional model acquisition and reconstruction system |
| CN113362384A (en) * | 2021-06-18 | 2021-09-07 | 安徽理工大学环境友好材料与职业健康研究院(芜湖) | High-precision industrial part measurement algorithm of multi-channel sub-pixel convolution neural network |
| CN114331842B (en) * | 2021-12-28 | 2024-04-05 | 武汉大学 | DEM super-resolution reconstruction method combining topographic features |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103390266A (en) * | 2013-07-31 | 2013-11-13 | 广东威创视讯科技股份有限公司 | Image super-resolution method and device |
| CN103455709A (en) * | 2013-07-31 | 2013-12-18 | 华中科技大学 | Super-resolution method and system for digital elevation model |
| CN106228512A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | Based on learning rate adaptive convolutional neural networks image super-resolution rebuilding method |
-
2016
- 2016-12-15 CN CN201611159517.0A patent/CN106600553B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103390266A (en) * | 2013-07-31 | 2013-11-13 | 广东威创视讯科技股份有限公司 | Image super-resolution method and device |
| CN103455709A (en) * | 2013-07-31 | 2013-12-18 | 华中科技大学 | Super-resolution method and system for digital elevation model |
| CN106228512A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | Based on learning rate adaptive convolutional neural networks image super-resolution rebuilding method |
Non-Patent Citations (2)
| Title |
|---|
| CONVOLUTIONAL NEURAL NETWORK BASED DEM SUPER RESOLUTION;Zixuan Chen等;《The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences》;20160630;第41卷(第B3期);摘要，第2-4页 * |
| Single Image Super-Resolution Based on Deep Learning and Gradient Transformation;Jingxu Chen等;《2016 IEEE 13th International Conference on Signal Processing(ICSP)》;20161110;摘要，第2-4页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106600553A (en) | 2017-04-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106600553B (en) | DEM super-resolution method based on convolutional neural network | |
| CN110738252B (en) | Space autocorrelation machine learning satellite precipitation data downscaling method and system | |
| CN108921799B (en) | Remote sensing image thin cloud removing method based on multi-scale collaborative learning convolutional neural network | |
| CN112836610B (en) | Land use change and carbon reserve quantitative estimation method based on remote sensing data | |
| CN110361778B (en) | Seismic data reconstruction method based on generation countermeasure network | |
| CN111553858B (en) | Image restoration method and system based on generation countermeasure network and application thereof | |
| CN111652892A (en) | Remote sensing image building vector extraction and optimization method based on deep learning | |
| DE102017006563A1 (en) | Image patch matching using probability based sampling based on prediction | |
| CN110400371B (en) | Three-dimensional model construction method for horizontally-constructed landform entity | |
| CN111340723B (en) | Terrain-adaptive airborne LiDAR point cloud regularization thin plate spline interpolation filtering method | |
| CN109389553B (en) | Meteorological facsimile picture contour interpolation method based on T spline | |
| CN111582093A (en) | Automatic small target detection method in high-resolution image based on computer vision and deep learning | |
| CN114331842B (en) | DEM super-resolution reconstruction method combining topographic features | |
| CN104299241A (en) | Remote sensing image significance target detection method and system based on Hadoop | |
| CN107563963B (en) | Super-resolution reconstruction method based on single depth map | |
| CN114067205A (en) | Light-weight arbitrary-scale double-time-phase image change detection method | |
| CN112001121B (en) | Large-area gentle region predictive soil mapping method based on solar radiation | |
| CN106897683B (en) | Ground object detection method and system of remote sensing image | |
| CN116883594A (en) | Ocean three-dimensional temperature remote sensing super-resolution reconstruction method integrating satellite and buoy observation | |
| CN114693755B (en) | Non-rigid registration method and system for multimode image maximum moment and space consistency | |
| CN116385892A (en) | Digital elevation model extraction method based on target context convolution neural network | |
| CN109856673B (en) | High-resolution Radon transformation data separation technology based on dominant frequency iterative weighting | |
| CN113627465A (en) | Rainfall data space-time dynamic fusion method based on convolution long-short term memory neural network | |
| CN112784433B (en) | Hierarchical simulated annealing modeling method based on corrosion | |
| Sayed et al. | An improved technique for LIDAR data reduction |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191217 Termination date: 20201215 |