Ship Detection from Optical Remote Sensing Image based on Size-Adapted CNN

Xin Hou, Qizhi Xu, Yan Ji

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

6 Citations (Scopus)

Abstract

Due to the diversity of ship sizes, ship detection from optical remote sensing images is still a challenging task. To tackle this problem, this paper proposed a size-adapted framework based on a coarse-to-fine strategy. First, the proposed method generates ship candidates through calculating the convolutional feature maps by three shallow convolutional neural networks (CNN) with different scales, in which the entire images with arbitrary sizes are fed as input; Second, a feature vector with fixed-length is extracted by the spatial pyramid pooling (SPP) layer from the candidates, regardless of the size and aspect ratio of the candidates; Finally, the multi-task learning is employed to classify the candidates using a softmax classifier, as well as to reduce the ship localization error by simple bounding-box regression. The experiments were carried out on various images, and the results demonstrated the effectiveness of the proposed method while dealing with the ships of different sizes.

Original languageEnglish
Title of host publication5th International Workshop on Earth Observation and Remote Sensing Applications, EORSA 2018 - Proceedings
EditorsQihao Weng, Paolo Gamba, Ni-Bin Chang, Guangxing Wang, Wanqiang Yao
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781538666425
DOIs
Publication statusPublished - 31 Dec 2018
Externally publishedYes
Event5th International Workshop on Earth Observation and Remote Sensing Applications, EORSA 2018 - Xi'an, China
Duration: 18 Jun 201820 Jun 2018

Publication series

Name5th International Workshop on Earth Observation and Remote Sensing Applications, EORSA 2018 - Proceedings

Conference

Conference5th International Workshop on Earth Observation and Remote Sensing Applications, EORSA 2018
Country/TerritoryChina
CityXi'an
Period18/06/1820/06/18

Keywords

  • multi-task learning
  • ship detection
  • size-adapted CNN
  • spatial pyramid pooling (SPP)

Fingerprint

Dive into the research topics of 'Ship Detection from Optical Remote Sensing Image based on Size-Adapted CNN'. Together they form a unique fingerprint.

Cite this