Aerial image segmentation method based on global and multi-scale full convolutional networks

A fully convolutional network and aerial image technology, which is applied in the field of aerial image segmentation, can solve the problems of not considering the different functions of the decoding part of the full convolution network, not considering the simplicity and efficiency of the network, and the low performance of network segmentation, so as to achieve robustness High, overcome simplicity and high efficiency, and reduce the effect of high-frequency information loss

Active Publication Date: 2020-02-28
XIDIAN UNIV
View PDF5 Cites 11 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that the convolutional encoding-decoding network structure in the convolutional encoding-decoding network relies on multi-layer convolutional layers. Due to the limitation of the size of the convolution kernel, only local information can be extracted, and there is a lack of global information, resulting in Segmentation accuracy is low
The shortcomings of this method are: only the relationship between the encoding and decoding parts of the full convolutional network is considered, the different effects of each convolutional layer on the final prediction in the decoding part of the full convolutional network are not considered, and the multi-scale features are not considered, resulting in It is difficult to identify similar objects of different sizes in the image, and the simplicity and efficiency of the network are not considered, resulting in low network segmentation performance

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
  • Aerial image segmentation method based on global and multi-scale full convolutional networks
  • Aerial image segmentation method based on global and multi-scale full convolutional networks
  • Aerial image segmentation method based on global and multi-scale full convolutional networks

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0037] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0038] Reference attached figure 1 , The implementation steps of the present invention will be further described in detail.

[0039] Step 1. Construct a global and multi-scale full convolutional network.

[0040] The first step is to build a global and multi-scale full convolutional network. Its structure is: input layer → feature extraction layer → first combination module → fully connected layer → deconvolution layer → second combination module → output layer.

[0041] The feature extraction layer is a VGG16 model composed of five convolution modules connected in series.

[0042] The first combination module has 7 layers, and its structure is in order: first convolution layer→transpose layer→first multiplication layer→softmax layer→second multiplication layer→second convolution layer→addition layer.

[0043] The structure of the fully connected layer is in order: maxi...

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 an aerial image segmentation method based on global and multi-scale full convolutional networks. The method comprises the following steps: constructing the global and multi-scale full convolutional networks; generating a training set; training global and multi-scale full convolutional networks; and inputting a to-be-segmented aerial image into the trained global and multi-scale full convolutional network for binary segmentation, and generating a segmentation mask image. According to the method, the aerial image is segmented by using the global and multi-scale full convolutional networks, and the global module and the multi-scale module are embedded into the global and multi-scale full convolutional networks, so that a finer segmentation mask is extracted, the robustness is high, and the segmentation precision is high.

Description

Technical field [0001] The present invention belongs to the field of image processing technology, and further relates to an aerial image segmentation method based on global and multi-scale full convolutional networks in the field of image segmentation technology. The invention can be used to detect the building target from the high-resolution aerial image and segment the building area from the image. Background technique [0002] With the continuous development of today's society, urban construction planning has become a hot topic of concern. With the increasing demand for buildings, more buildings have increased the difficulty of urban infrastructure construction, such as transportation route planning, drainage system planning, and convenience facilities planning. Building detection and segmentation in aerial images can help the construction planning department to detect and segment urban buildings and carry out urbanization infrastructure construction. However, due to the ric...

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 Applications(China)
IPC IPC(8): G06T7/11G06N3/04
CPCG06T7/11G06T2207/20081G06T2207/20084G06N3/045
Inventor 马晶晶吴琳琳唐旭焦李成
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