Land Cover Classification from Satellite Imagery with U-Net and Lovász-Softmax Loss

The land cover classification task of the DeepGlobe Challenge presents significant obstacles even to state of the art segmentation models due to a small amount of data, incomplete and sometimes incorrect labeling, and highly imbalanced classes. In this work, we show an approach based on the U-Net ar...

Full description

Saved in:
Bibliographic Details
Main Authors: Rakhlin, Alexander, Davydow, Alex, Nikolenko, Sergey
Format: Conference Proceeding
Language:English
Subjects:
Computer architecture
Computer vision
Image segmentation
Satellites
Stochastic processes
Task analysis
Training
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The land cover classification task of the DeepGlobe Challenge presents significant obstacles even to state of the art segmentation models due to a small amount of data, incomplete and sometimes incorrect labeling, and highly imbalanced classes. In this work, we show an approach based on the U-Net architecture with the Lov´asz-Softmax loss that successfully alleviates these problems; we compare several different convolutional architectures for U-Net encoders.
ISSN:2160-7516
DOI:10.1109/CVPRW.2018.00048