VEHICLE DETECTION IN SATELLITE IMAGES WITH DEEP NEURAL NETWORKS AND VEHICLE SHAPE FEATURES

Kaiji He, Long Zhang

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

Abstract

The vehicle geographical distribution information including
vehicle location and quantity is essential for planning the future
power network capacity as all the conventional vehicles
will be replaced by the electric vehicles (EVs) in the next
few decades. However, collecting such information manually
is expensive and time-consuming. With the development of
computer vision techniques, it brings an opportunity to build
an intelligent detector that can detect vehicles in satellite images
automatically, which can help the power energy suppliers
gather vehicle information more efficiently. In this paper,
two convolutional neural networks (CNN), SegNet and Mask
R-CNN, are employed for identifying vehicles from satellite
images. The CNN results are further improved by support
vector machine (SVM) with vehicle shape features. The real
satellite images covering central Manchester area are used to
demonstrate the effectiveness of the proposed methods.
Original languageEnglish
Title of host publicationInternational Geoscience and Remote Sensing Symposium
Publication statusPublished - 2021

Keywords

  • deep convolutional neural networks
  • electric vehicle detection
  • satellite images

Fingerprint

Dive into the research topics of 'VEHICLE DETECTION IN SATELLITE IMAGES WITH DEEP NEURAL NETWORKS AND VEHICLE SHAPE FEATURES'. Together they form a unique fingerprint.

Cite this