skip to main content

DOE PAGESDOE PAGES

Title: A mixed-scale dense convolutional neural network for image analysis

We report that deep convolutional neural networks have been successfully applied to many image-processing problems in recent works. Popular network architectures often add additional operations and connections to the standard architecture to enable training deeper networks. To achieve accurate results in practice, a large number of trainable parameters are often required. Here, we introduce a network architecture based on using dilated convolutions to capture features at different image scales and densely connecting all feature maps with each other. The resulting architecture is able to achieve accurate results with relatively few parameters and consists of a single set of operations, making it easier to implement, train, and apply in practice, and automatically adapts to different problems. Lastly, we compare results of the proposed network architecture with popular existing architectures for several segmentation problems, showing that the proposed architecture is able to achieve accurate results with fewer parameters, with a reduced risk of overfitting the training data.
Authors:
 [1] ;  [2]
  1. Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States)
  2. Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States); Univ. of California, Berkeley, CA (United States)
Publication Date:
Grant/Contract Number:
AC02-05CH11231; AC03-76SF00098
Type:
Published Article
Journal Name:
Proceedings of the National Academy of Sciences of the United States of America
Additional Journal Information:
Journal Volume: 115; Journal Issue: 2; Journal ID: ISSN 0027-8424
Publisher:
National Academy of Sciences, Washington, DC (United States)
Research Org:
Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States)
Sponsoring Org:
USDOE Office of Science (SC), Advanced Scientific Computing Research (ASCR) (SC-21); USDOE Office of Science (SC), Basic Energy Sciences (BES) (SC-22); USDOE Office of Science (SC), Biological and Environmental Research (BER) (SC-23)
Country of Publication:
United States
Language:
English
Subject:
97 MATHEMATICS AND COMPUTING; image segmentation; machine learning; convolution neural networks
OSTI Identifier:
1414877
Alternate Identifier(s):
OSTI ID: 1485062

Pelt, Daniël M., and Sethian, James A.. A mixed-scale dense convolutional neural network for image analysis. United States: N. p., Web. doi:10.1073/pnas.1715832114.
Pelt, Daniël M., & Sethian, James A.. A mixed-scale dense convolutional neural network for image analysis. United States. doi:10.1073/pnas.1715832114.
Pelt, Daniël M., and Sethian, James A.. 2017. "A mixed-scale dense convolutional neural network for image analysis". United States. doi:10.1073/pnas.1715832114.
@article{osti_1414877,
title = {A mixed-scale dense convolutional neural network for image analysis},
author = {Pelt, Daniël M. and Sethian, James A.},
abstractNote = {We report that deep convolutional neural networks have been successfully applied to many image-processing problems in recent works. Popular network architectures often add additional operations and connections to the standard architecture to enable training deeper networks. To achieve accurate results in practice, a large number of trainable parameters are often required. Here, we introduce a network architecture based on using dilated convolutions to capture features at different image scales and densely connecting all feature maps with each other. The resulting architecture is able to achieve accurate results with relatively few parameters and consists of a single set of operations, making it easier to implement, train, and apply in practice, and automatically adapts to different problems. Lastly, we compare results of the proposed network architecture with popular existing architectures for several segmentation problems, showing that the proposed architecture is able to achieve accurate results with fewer parameters, with a reduced risk of overfitting the training data.},
doi = {10.1073/pnas.1715832114},
journal = {Proceedings of the National Academy of Sciences of the United States of America},
number = 2,
volume = 115,
place = {United States},
year = {2017},
month = {12}
}