Hyperspectral and Panchromatic Image Fusion Method Based on Spatial Features Extracted by Aae

A hyperspectral image and panchromatic image technology, applied in image enhancement, image analysis, graphic image conversion, etc., can solve the problems of sub-optimal, imperfect spectral information, and high algorithm complexity

Active Publication Date: 2021-09-10
XIDIAN UNIV
View PDF7 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that the spectral information preservation is constrained by using the output of the spatial preservation network and the original image, that is, the spectral information preservation depends on the spatial information preservation, which is indirect and may lead to suboptimal preservation results.
Although this method obtains a fusion result with improved resolution by training the neural network, the method still has the disadvantage that it does not maintain the spectral information and spatial information well, and it is optimized by using an adaptive moment estimation algorithm. Due to the huge amount of hyperspectral image data, and the sharpening process requires the entire network to participate in training, the calculation process is very cumbersome, and the algorithm complexity is high, which lacks the combination of traditional image processing methods and neural network-based methods, and the spectral information is also maintained. not perfect

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Hyperspectral and Panchromatic Image Fusion Method Based on Spatial Features Extracted by Aae
  • Hyperspectral and Panchromatic Image Fusion Method Based on Spatial Features Extracted by Aae
  • Hyperspectral and Panchromatic Image Fusion Method Based on Spatial Features Extracted by Aae

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0040] Due to the limitation of incident light energy, there is a certain balance between the spectral resolution, spatial resolution and signal-to-noise ratio of remote sensing data obtained by different sensors. Hyperspectral imaging sensors can acquire hyperspectral images with hundreds of narrow spectral channels and have high spectral resolution. Since hyperspectral images have detailed spectral information, they have unique advantages in classification, detection, and identification. However, the spatial resolution of hyperspectral images is insufficient. When processing and utilizing images, it is necessary to obtain hyperspectral images with both high spatial and spectral resolutions.

[0041] Using neural networks for image fusion is usually to learn the mapping relationship between input images and reconstructed fusion images. It has great advantages in feature extraction, data representation and description of complex relationships, but the design of neural network ...

Embodiment 2

[0076] The hyperspectral and panchromatic image fusion method based on extracting spatial features against an autoencoder is the same as in Embodiment 1, and the training process of the network described in step 3 of the present invention includes the following steps:

[0077] The adversarial autoencoder can be divided into two major network structures: the autoencoder part and the discriminative part of the generative adversarial network (GAN). After initializing the weights and biases of the adversarial autoencoder, training the adversarial autoencoder is divided into two processes:

[0078] (3.1) Reconstruction of input samples: the upsampled hyperspectral image H is used as the input of the adversarial autoencoder for training, and the loss function of the autoencoder is calculated; the stochastic gradient descent algorithm is used for optimization, and the optimization process is to minimize the loss function The process of optimizing, while updating the network weights a...

Embodiment 3

[0082] The hyperspectral and panchromatic image fusion method based on extracting spatial features against the self-encoder is the same as that in Embodiment 1-2, and the panchromatic image is enhanced as described in step 4, specifically including the following steps:

[0083] (4.1) Perform adaptive histogram equalization on the panchromatic image P:

[0084] P h =adapt thisteq(P)

[0085] Among them, P represents a panchromatic image, adaptthisteq( ) represents an adaptive histogram equalization function, and P h Represents a panchromatic image after adaptive histogram equalization;

[0086] (4.2) Use the Laplacian-Gaussian (LOG) enhancement algorithm to increase the panchromatic image P after adaptive histogram equalization h space details.

[0087] (4.2.1) Use the Gaussian filter to remove the noise of the panchromatic image after adaptive histogram equalization:

[0088] P g =P h × g

[0089] Among them, P h Represents the panchromatic image after adaptive histog...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a hyperspectral and panchromatic image fusion method based on AAE extraction of spatial features, which solves the problems of serious spectral distortion and insufficient injection of spatial details in existing hyperspectral image fusion. Processing; building and training hyperspectral image spectral constraints against self-encoder network; extracting spatial features of hyperspectral image; obtaining spatial details of enhanced panchromatic image; fusing spatial information of hyperspectral image and enhanced panchromatic image; building gain matrix; get a hyperspectral image with high spatial resolution. The invention considers hyperspectral and panchromatic image space information at the same time, uses AAE to extract deep spatial features, effectively reduces spectral loss, and effectively improves spatial resolution. The simulation proves that the fusion index and visual effect space and spectrum performance of the present invention are better, the spatial details are kept better, and the spectral information is more complete, which is used for hyperspectral image fusion processing.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to fusion of hyperspectral images, in particular to a fusion method of hyperspectral images and panchromatic images based on extraction of spatial features by an adversarial autoencoder (AAE). The invention can be used to obtain hyperspectral images with high spatial resolution and rich spectral information. Background technique [0002] Images with high spatial resolution are conducive to precise positioning of targets, and images with high spectral resolution are conducive to distinguishing categories of different features and accurately identifying targets. Hyperspectral images have more bands in a certain wavelength range and high spectral resolution. They have been used in many fields and some practical applications, such as vegetation research, precision agriculture, regional geological filling images, mineral exploration and environmental monitoring, etc. . L...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Patents(China)
IPC IPC(8): G06T5/50G06T3/40G06K9/62G06N3/04G06N3/08
CPCG06T5/50G06T3/4023G06T3/4076G06N3/08G06T2207/10036G06N3/045G06F18/253
Inventor 谢卫莹钟佳平李云松雷杰刘保珠
Owner XIDIAN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

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