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

Outdoor scene multiobjective segmentation method based on depth convolutional neural network

A deep convolution, neural network technology, applied in image analysis, image data processing, instruments, etc., can solve the problem of image prediction results prone to mismatch, wrong segmentation results, and inaccurate segmentation results.

Active Publication Date: 2018-01-09
UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF3 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] 1) Image prediction results are prone to mis-match
[0005] 2) The segmentation result is wrong when the image contains confusing categories
[0006] 3) The segmentation results are inaccurate when the image contains inconspicuous categories

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
  • Outdoor scene multiobjective segmentation method based on depth convolutional neural network
  • Outdoor scene multiobjective segmentation method based on depth convolutional neural network
  • Outdoor scene multiobjective segmentation method based on depth convolutional neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0017] The implementation process is as follows figure 1 As shown, the steps are as follows:

[0018] Step 1: Decentralize the input original image (size 480*480). According to the pre-calculated mean values ​​of the three RGB channels of the image in the training database, they are 104.008, 116.669, and 122.675, respectively. The corresponding mean value is subtracted from the three channels of each input image, which can make the model run more stably.

[0019] Step 2: The feature extraction module uses a combination of 13 convolutional layers and 4 pooling pool layers to obtain 4 feature spectra of different scales. The size is: 240*240*128 (height*width*number of channels) , 120*120*256, 60*60*512, 30*30*512. The convolution layer uses a filter with a kernel size of 3*3 and a step size of 1. The number of filters starts from the bottom layer output, increases along with the deepening of the number of layers, and takes a value of 64, 128, 256, 512 (13 convolutional laye...

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 provides an outdoor scene multiobjective segmentation method based on a depth convolutional neural network. The method comprises the steps of feature extraction, feature fusion, multi-scale pooling and upsampling. An upsampling network consists of two convolutional layers and a data transformation layer. The later output characteristic spectra of two convolution layers pass through the data transformation layer to acquire a characteristic spectrum which is restored to the size of an original input image. According to the invention, sufficient context information is introduced through characteristic fusion on different characteristic spectra, and multi-scale pooling operation is carried out on the fused characteristic spectrum to acquire global information in different receptive fields.

Description

technical field [0001] The invention relates to image segmentation technology. Background technique [0002] At this stage, research on deep learning technology in the fields of driverless cars, smart car technology, and smart mobile robots is becoming more and more popular, and target scene segmentation technology is one of the key points in the above fields. The scene segmentation technology perceives and understands various scenes, and outputs the segmentation results to the navigation system or path planning system, so as to guide the mobile device to take the next action. [0003] Image segmentation is a crucial preprocessing for image recognition and computer vision. When the segmentation is automatically processed by a computer, various difficulties will be encountered. For example, uneven lighting, the influence of noise, unclear parts in the image, and shadows often cause segmentation errors. Existing segmentation algorithms based on machine learning rely too much...

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): G06T7/10G06T3/40G06K9/62
Inventor 李宏亮张文海翁爽董蒙孙玲
Owner UNIV OF ELECTRONICS SCI & TECH OF CHINA