PARALLEL IMPLEMENTATION OF THE FLICM ALGORITHM FOR SAR IMAGE
CHANGE DETECTION ON INTEL MIC
Huming Zhu, Le Lu, Yucong Fan, Pei Li, Qingyu Zhang, Licheng Jiao
Key Laboratory of Intelligent Perception and Image Understanding of Ministry of Education,
International Research Center for Intelligent Perception and Computation,
Xidian University, Xi’an, Shaanxi Province710071, China
ABSTRACT
Image change detection has a wide range of applications in
various fields, such as damage assessment, environmental
monitoring and agricultural surveys. As the number of
remote sensing images and the complexity of algorithm rise,
the demand for processing power is increasing. In this paper,
we present a parallel FLICM algorithm for SAR image
change detection on Intel MIC (Many Integrated Core)
which is the new type of many-core coprocessor. The
proposed algorithm is implemented based on MIC-Offload
mode using OpenMP (Open Multi-Processing). The parallel
characteristics and implementation details of the proposed
parallel FLICM algorithm are presented. Experiment results
demonstrate that the optimized parallel algorithm can
greatly reduce the computational time of the change
detection algorithm. The speedup is up to 20× compared
with the runtime of serial algorithm on CPU.
Index Terms—Change detection, FLICM algorithm, MIC,
OpenMP
1. INTRODUCTION
SAR image change detection is to detect the changes of the
image in the same area taken at different times. It has a wide
range of applications in various fields, such as the change of
forest coverage, urban expansion, land use, the growth
status of crop, etc. In the last few years, a large amount of
images have been acquired by remote sensing sensors over
the earth, their size is increasing rapidly with the increment
of image resolution and the number of satellites. Moreover,
some of them are freely available for the scientific
community (e.g., the entire images from Landsat, the
Sentinel 1A images). These changes mentioned above make
traditional methods of image change detection more and
more inefficient and powerless for larger scale images. It
has become a new challenge that how to deal with the SAR
images quickly.
Recently, a variety of methods of unsupervised SAR
image change detection are proposed. M. Gong et al. [1]
present a novel approach for change detection in SAR
images which classifies changed and unchanged regions by
fuzzy c-means clustering with a novel Markov random field
(MRF) energy function. A. Garzelli et al. [2] propose an
unsupervised nonparametric method for change detection in
multitemporal SAR images, etc. Nowadays, there are a
variety of heterogeneous computing devices appearing, such
as GPU, FPGA and DSP, which could accelerate the image
processing of remote sensing. For example, M. Hossain et
al.[3] present a novel technique for FPGA based
implementation of high resolution SAR system using UWB-
OFDM architecture. H. Zhu et al. [4] accelerate an
unsupervised cluster algorithm: Fuzzy Local Information C-
means (FLICM) clustering algorithm on GPU, which has
achieved 100× speedup approximately.
In recent years, MIC is a new kind of coprocessor
architecture designed by Intel Corporation. It has more cores
and wider vector units which make the platform more
efficient in improving performance and meet the need of
highly parallel computing applications when compared with
multi-core CPU. The parallel algorithm on MIC has
advantages in extending to different heterogeneous platform
on the basis of reducing the time consumption when
compared with GPU. Nowadays, Intel MIC has been applied
in remote sensing. For example, F. Serra Pajuelo et al. [5]
present several optimizations for hyperspectral image
processing algorithms which can be successfully
implemented on Intel MIC. J. Mielikainen et al. [6] propose
the optimization results for the Purdue-Lin microphysics
scheme which runs natively on Intel Xeon Phi.
With the increasing scale of SAR image, traditional
serial algorithms have shown their low efficiency when
dealing with SAR image change detection. It is necessary to
change the traditional serial algorithm to parallel algorithm
in order to improve efficiency. In this paper, we present
parallel unsupervised SAR image change detection based on
FLICM algorithm on Intel MIC.
2. MICARCHITECTURE
The MIC coprocessor is consisting of a number of high-
speed interconnected processor cores. In addition to the Intel
Architecture (IA) computing cores, the MIC coprocessor has
8 memory controllers, and supports 16 GDDR5 transmission
2340978-1-5090-3332-4/16/$31.00 ©2016 IEEE IGARSS 2016