Heterogeneous image change detection method based on non-supervision depth neural network
A deep neural network and image change detection technology, applied in the field of remote sensing image processing, can solve problems such as complex implementation and inability to meet requirements, and achieve excellent feature learning ability, excellent effect, and stable results.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Example Embodiment
[0041] Example 1:
[0042] This embodiment provides a heterogeneous image change detection method based on an unsupervised deep neural network, such as figure 1 As shown, including the following steps:
[0043] Step 1: Select two heterogeneous images of the same area in different time phases and mark them as image I 1 And image I 2 , Using deep neural networks to image I 1 The neighborhood information of all points is input, reconstructed image I 2 The neighborhood information of, get the initial reconstruction mapping function f 1 (x), get the initial difference map DI 1 ;
[0044] Step 2: The initial difference map DI obtained in step 1 1 Select sample points in, retrain the deep neural network, and get the final reconstruction mapping function f(x);
[0045] Step 3: Use the final reconstruction mapping function f(x) obtained in Step 2 to obtain the difference map DI, and obtain the final change detection result.
[0046] The present invention breaks through the traditional heterogen...
Example Embodiment
[0047] Example 2:
[0048] This embodiment further describes step one in detail on the basis of embodiment 1, as figure 2 As shown, step one specifically includes the following steps:
[0049] Step 101: Select two heterogeneous images of the same area in different time phases, and mark them as image I 1 And image I 2 , Take the pixel at position (i, j) as the central pixel, take a window with a size of 5×5, and the total number of pixels is N=25, extract two images I 1 , I 2 Neighborhood information IF 1 , IF 2 ;
[0050] Step 102: Initialize the deep neural network randomly, initialize the iteration counter t=0, and the maximum number of iterations T=50;
[0051] Step 103: The image I 1 Neighborhood information IF 1 Input point by point into the deep neural network of step 102, and take the image I 2 Neighborhood information IF 2 The corresponding point of is used as the label, and the network parameters are updated using the conjugate gradient algorithm based on the minimum cross en...
Example Embodiment
[0055] Example 3:
[0056] This embodiment further describes step two in detail on the basis of embodiment 1 and embodiment 2, as image 3 As shown, step two specifically includes the following steps:
[0057] Step 201: Convert the initial difference map DI obtained in step 1 1 Perform threshold segmentation to get the initial detection result Iout 1 ;
[0058] Step 202: Select IF 1 , IF 2 Corresponding Iout in 1 Points with 0 in the middle form a new set IF 10 , IF 20 ;
[0059] Step 203: initialize the deep neural network randomly, initialize the iteration counter t=0, and the maximum number of iterations T=50;
[0060] Step 204: Set the IF 10 Input point by point into the deep neural network of step 203, and use IF 20 The corresponding point of is used as the label, and the network parameters are updated using the conjugate gradient algorithm based on the minimum cross entropy, t=t+1;
[0061] Step 205: Step 204 is continuously repeated until the error is less than a prescribed thresh...
PUM
Abstract
Description
Claims
Application Information
- R&D Engineer
- R&D Manager
- IP Professional
- Industry Leading Data Capabilities
- Powerful AI technology
- Patent DNA Extraction
Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.
© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap