Vous êtes sur la page 1sur 2

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/272291633

Efficient Framework for Palm Tree Detection in UAV Images

Article  in  IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · December 2014
DOI: 10.1109/JSTARS.2014.2331425

CITATIONS READS
59 1,069

5 authors, including:

Salim Malek Yakoub Bazi


Università degli Studi di Trento King Saud University
13 PUBLICATIONS   302 CITATIONS    97 PUBLICATIONS   3,373 CITATIONS   

SEE PROFILE SEE PROFILE

Naif Alajlan Haikel Salem Alhichri


King Saud University King Saud University
120 PUBLICATIONS   2,401 CITATIONS    50 PUBLICATIONS   913 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Landscape Archaeology in Northern Africa View project

Remote Sensing Scene Classification View project

All content following this page was uploaded by Yakoub Bazi on 18 February 2015.

The user has requested enhancement of the downloaded file.


4692 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 7, NO. 12, DECEMBER 2014

Efficient Framework for Palm Tree Detection


in UAV Images
Salim Malek, IEEE Member, Yakoub Bazi, Senior Member, IEEE, Naif Alajlan, Senior Member, IEEE,
Haikel AlHichri, IEEE Member, and Farid Melgani, Senior Member, IEEE

Abstract—The latest developments in unmanned aerial vehicles to advancement in technologies and reduction in prices, UAVs
(UAVs) and associated sensing systems make these platforms became a practical solution for many civilian applications
increasingly attractive to the remote sensing community. The such as vegetation monitoring [6], [7] anomaly detection in
large amount of spatial details contained in these images opens
the door for advanced monitoring applications. In this paper, we archeological sites [8], precision farming [10]–[12], and car
use this cost-effective and attractive technology for the automatic counting [13]. The high flexibility in the use of small UAV sys-
detection of palm trees. Given a UAV image acquired over a palm tems permits immediate intervention and interactive measure-
farm, first we extract a set of keypoints using the Scale-invariant ments according to customer’s specific needs. In fact, UAVs:
Feature Transform (SIFT). Then, we analyze these keypoints with 1) allow mapping and monitoring small areas at extremely fine
an extreme learning machine (ELM) classifier a priori trained
on a set of palm and no-palm keypoints. As output, the ELM scales (up to 1 cm); 2) can reach the desired point of obser-
classifier will mark each detected palm tree by several keypoints. vation in just few minutes; and 3) allow multitemporal acqui-
Then, in order to capture the shape of each tree, we propose sitions over the same area at predefined and desired times.
to merge these keypoints with an active contour method based However, on the other hand the UAV images introduce extra
on level sets (LSs). Finally, we further analyze the texture of challenges compared to VHR satellite images as they are char-
the regions obtained by LS with local binary patterns (LBPs)
to distinguish palm trees from other vegetations. Experimental acterized by improved spatial resolutions making traditional
results obtained on UAV images with 3.5 cm of spatial resolution remote sensing image analysis methods not well adapted.
and acquired over two different farms confirm the promising In this paper, we propose to exploit this technology for
capabilities of the proposed framework. detecting date palm trees in farms situated in the kingdom
Index Terms—Extreme learning machine (ELM), level set (LS), of Saudi Arabia. The interest in obtaining such information
local binary pattern (LBP), palm trees, scale-invariant feature is explained by several reasons. First, the date palm is one
transform (SIFT), unmanned aerial vehicle (UAV) images. of the most important crops in the Kingdom as the country
accounts around 23 millions of trees [14]. Second, it provides
I. I NTRODUCTION an inventory of the trees that may help in a better planning of
the irrigation process. Third, the information about the exact
I N RECENT years, very high resolution (VHR) images
with spatial resolutions less than 1 m are constantly
acquired over the earth surface by VHR satellites such as
number of trees is essential for predicting the date production.
Finally, the number of trees is used as a fundamental criterion
GeoEye-1 and Worldview-2. The advent of this type of im- for obtaining public grants. Usually, the process of counting
agery opened the door for new applications such as counting palm trees is carried out manually on site. Such an operation
roofless buildings [1], object tracking and identification [2], is very tedious and time-consuming. Therefore, the utilization
[3], quantifying bird migration [4], and automated detection of automatic methods represents for the public and private
of buildings [5]. agricultural institutions a potentially interesting alternative.
Besides VHR satellites, unmanned aerial vehicles (UAVs) From a methodological point of view, we present in this
are also increasingly adopted by the remote sensing commu- paper an automatic framework composed of four main phases.
nity as new effective acquisition systems. For a long time, Given a UAV image acquired over a palm farm, we first extract
UAVs were used exclusively for military applications, but due a set of candidate keypoints using the Scale-invariant Feature
Transform (SIFT) [15]. Then, we analyze these keypoints with
Manuscript received April 01, 2014; accepted June 05, 2014. Date of a recent kernel-based classification method termed as extreme
publication August 20, 2014; date of current version January 21, 2015.
This work was supported by NSTIP strategic technologies programs under learning machine (ELM) [16]–[18]. This last is a priori trained
10-SPA1192-02 in Saudi Arabia. on palm and no-palm keypoints extracted by SIFT from a
S. Malek is with Department of Electronic, Saad Dahleb University, Blida, set of a palm and no-palm training templates. As output, the
Algeria (e-mail: smalek@ksu.edu.sa).
Y. Bazi, N. Alajlan, and H. AlHichri are with ALISR Laboratory, ELM classifier will mark each detected palm tree by several
College of Computer and Information Sciences, King Saud University, keypoints. Then, in order to capture the shape of each tree,
Riyadh 11543, Saudi Arabia (e-mail: ybazi@ksu.edu.sa; najlan@ksu.edu.sa; we propose to merge these keypoints with an active contour
hhichri@ksu.edu.sa).
F. Melgani is with the Department of Information Engineering and Compu- method based on level sets (LSs). Because of the keypoints
ter Science, University of Trento, I-38123 Trento, Italy (e-mail: melgani@disi. incorrectly classified by ELM as palm, side effects could
unitn.it). appear at the end of the grouping process such as merging very
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. close trees and the creation of regions related to other green
Digital Object Identifier 10.1109/JSTARS.2014.2331425 vegetation. To this end, we introduce processing operations

1939-1404 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

View publication stats

Vous aimerez peut-être aussi