Unlock instant, AI-driven research and patent intelligence for your innovation.

A Semantic Segmentation Method of Multispectral Image Based on Convolutional Neural Network

A convolutional neural network and multi-spectral image technology, which is applied in biological neural network models, image analysis, image enhancement, etc., can solve the problems of loss of computing time, loss of high-resolution image space information, interference image cutting, etc., to achieve The effect of improving precision, improving work efficiency and ensuring precision

Active Publication Date: 2022-04-22
THE THIRD RES INST OF CHINA ELECTRONICS TECH GRP CORP
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

If all low-resolution images are forcibly interpolated and enlarged to unify them with high-resolution images, some convolution operations on the low-resolution image parts will be invalid, not only losing a lot of computing time, but also possibly interfering with The result of image cutting
If the high-resolution image is down-sampled to make it consistent with the low-resolution image, a large amount of spatial information of the high-resolution image will be lost.

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
  • A Semantic Segmentation Method of Multispectral Image Based on Convolutional Neural Network
  • A Semantic Segmentation Method of Multispectral Image Based on Convolutional Neural Network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0022] For multispectral images, the bands are first separated according to the wavelength, and then the independent convolution operation is performed on different bands, that is, the convolutional neural network is used to convolve each data channel of the multispectral image independently, and then the feature map after the independent convolution of each data channel is fused (concatenation, summation). When convolving each data channel of a multispectral image independently, different sizes and numbers of convolutional nuclei are selected according to different bands. When convolving each data channel of a multispectral image independently, different convolutional layers are selected according to different bands. When implemented, convolutional neural networks employ U-NET neural networks.

Embodiment 2

[0024] When the multispectral image has a variety of resolutions, in the embodiment of example one using multi-channel independent convolution, the implementation of two using multi-channel independent convolution, multi-resolution input network. as Figure 2 As shown, the U-NET network is transformed into a convolutional neural network that supports multiple resolution inputs. Similar to conventional U-NET networks, the network of the present invention is composed of a scale shrinkage portion and a scale expansion portion, the scale shrinkage portion consists of a classic convolutional network, with the increase of the level of convolution, the image size decreases with the increase in the number of convolutional pooling, and the number of convolutional kernels increases with the increase in the number of pooling. The scale-expanded part is the same as the scale-expanding part of a U-NET network, and for each upsampling step of the scale-expanded part, the scale is tripled and the...

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 relates to a multi-spectral image semantic cutting method based on a convolutional neural network. The convolutional neural network is used to independently convolve each data channel of the multi-spectral image, and then independently convolve the feature map of each data channel. Perform fusion. The invention effectively solves the problem that the standard U‑NET network can only accept one RGB\Gray image of the same scale through a multi-resolution input and multi-channel independent convolution network, and effectively improves the working efficiency of multi-spectral image semantic cutting. Ensure the accuracy of image cutting.

Description

Technical field [0001] The present invention relates to a multispectral image semantic cutting method based on a convolutional neural network. Background [0002] Currently, advanced semantic cutting frameworks for RGB images generally employ end-to-end deep convolutional neural networks (DCNNs). The current use of convolutional neural networks is to classify objects using some pre-trained models, which are mainly VGG, ResNet and so on. The DCNN for the purpose of semantic cutting often includes two parts, the first half is the commonly used good quality DCN network, and the second half is the network that maps the feature map to pixel tags. In order to save training samples, the first half directly uses the pre-trained model parameters, and only the second half of the model parameters are fine-tuned. [0003] The current representative image semantic cutting network is the fully-convolutional network FCN, the initial version of which was based on VGG-16. Due to the fully connec...

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): G06T7/10G06N3/04
CPCG06T7/10G06T2207/20084G06T2207/10036G06N3/045
Inventor 李含伦戴玉成张小博张晓灿唐文
Owner THE THIRD RES INST OF CHINA ELECTRONICS TECH GRP CORP