209 
ASAR IMAGE TARGET RECOGNITION BASED ON THE COMBINED WAVELET 
TRANSFORMATION 
HE Hui, PENG Wanglu 
College of Information Technology and Software Engineering, Beijing Normal University at Zhuhai, Zhuhai, China 
Commission VII, WG VII/2 
KEY WORDS: Advanced Synthetic Aperture Radar, Target Recognition, Combined Wavelet Transformation, Cloud Model, 
Mathematical Morphology 
ABSTRACT: 
ASAR image target detection and recognition has been always a hot research. In this paper, on the foundation of the previous studies, 
a target recognition method based on combined wavelet transformation and cloud model is proposed. According to the good local 
characteristic, multi-resolution, effectiveness and the sensitivity to both direction and texture of the combined wavelet 
transformation, the detection method based on combined wavelet restrained the image noise and the errors, such as the object 
missing and false warning, caused by the use of the detection method based on the brightness alone. Then in order to solve the 
uncertainty of pixels, especially the object edge, an integrated method named soft segmentation for image segmentation based on the 
linguistic cloud model, which is a model for the conversion between qualitative and quantitative in the field of artificial intelligence 
is applied. Accurate target recognition is achieved after further processing eventually, such as mathematical morphology analysis. 
This integrated mechanism is applied to the ASAR data acquired from Zhuhai, China for carrying on the fish ponds recognition as 
well as the comparative experiments with classical methods. Results show that this approach can recognize the target more 
accurately and quick, which indicate that the synthetic scheme for target detection and recognition is flexible and robust and the 
advantages to the traditional detection operators or crisp segmentation methods are distinct. 
1. INTRODUCTION 
Since 1970's, various countries, such as America and England, 
have started to carry on the research on the automatic 
processing of SAR. One of the typical examples is extracting 
some significant characteristics from the SAR Imagery, such as 
region division, target detection and recognition, edge 
extraction and so on, which are impelled by the rapid 
development of computer technology. In addition, with the 
increasing development of radar signal processing technology, 
the SAR imagery resolution has been largely enhanced, which 
enables the realization for the automatic target detection of 
SAR Imagery. ASAR (Advanced Synthetic Aperture Radar) has 
been the most advanced imaging sensor on ENVISAT - 1 till 
now, which works in the C band with five kind of imaging 
patterns, seven kind of imaging strips and alternate polarized 
imaging function. Furthermore, besides all-weather, all-day and 
the certain penetrating characteristics like SAR, its data have 
unique advantages to any other radar sensors ( Qingni, 2004). 
Target detection and recognition on radar imagery is always the 
hot but challenging issue during the latest decades, tackled by a 
series of successful methods (.Lopes etc., 1993; Oliver etc., 
1996; Oliver, 1994). However, the radar remote sensing 
imagery detection not only involves the pure target detection 
simply, but also involves other instances, as the existence of 
speckles can also affect the target detection and recognition. In 
that case, it is needed to suppress the speckles to enhance the 
recognition accuracy. In the scope of the intensity of echoes, 
targets may be divided into strong and weak reflecting targets. 
The weak reflecting targets like fish ponds should belong to the 
category of target recognition 
The wavelet transformation has been one of the applied 
mathematical branches since later 1980s. Because of its partial, 
multi-resolution analysis characteristics, the wavelet 
transformation has already become one of the powerful image 
signal analysis tools (Szu etc. 2002).The wavelet transformation 
has also been widely applied to the SAR imagery target 
detection (Zhengjun, 1999; Ling etc. 2004; Jie etc. 2003). 
Besides, there are uncertainty factors of the remote sensing 
information (Yong, etc. 2004), the traditional crisp 
segmentation algorithm, can not describe images of uncertainty, 
especially for the edge pixels. The cloud model brought forward 
by LI Deyi (Deyi etc., 1998) can be used for the characteristics 
of a simple and accurate mathematical description of the 
concept of uncertainty, provides a quantitative and qualitative 
transformation tools that can effectively achieve the image of 
the soft partition. 
Series of experiments about fish ponds recognition are carried 
out in this article. To enhance the target recognition accuracy 
and efficiency, according to the previous studies, a new 
comprehensive recognition mechanism based on the combined 
wavelet transformation and linguistic cloud model is proposed, 
-in this article, experiments on the ASAR imagery of 
Guangdong, Zhuhai are presented both to validate the 
feasibility of the comprehensive mechanism and to compare 
with the traditional methods. 
2. FRAMEWORK OF COMBINED WAVELET 
In order to collect the edge information and partial texture 
information to suppress the disturbance of target detection, a 
framework of combined wavelets is proposed, one of which is