and light energy e when assumed that other imaging 
parameters are constant, which described as following, 
g = H(A,e) (l) 
Çj)r dach specific type of foreign fibersgthe' optimal wavelength 
°P‘ anithé optimal light energy opt for the maximum 
contrast 5 opt between foreign fibers and cotton are achieved 
in the following equation, 
g, P ,‘=H(Âj,e op ;) (2) 
Thus, we can define the OOP (optimal operating point) 
including the optimal wavelength and the optimal light energy 
for any type of detected foreign fibers, where image features of 
foreign fibers are most distinct in the spectral imaging. 
At different optimal operating point, we can capture the 
KÂ, ' c 
different optimal image opl ’ npl for respectively 
detected foreign fibers. And, given the number of the types of 
foreign fibers is ^ , it is reasonable to hypothesis that we can 
get the optimal image group ^ at ^ different OOP. 
AV’O) < 3 > 
Then the fused information from different optimal images was 
considered, which provide complete information for a wide 
range of foreign fibers. Also, the detailed features of foreign 
fibers were enhanced using image fusion method. In this paper, 
the wavelet-based image fusion approaches was described as 
following, 
where > I op! represents I(A op ‘,e op ‘) respectively, 
and F wavelgt is function of wavelet images fusion. 
Thus, a multiwavelength image collection system can be 
implemented by the images fused method, according which 
each acquired image at the optimal operating point was fused 
into the single image, so the fused feature in the single image 
can greatly enhance our capability to identify the foreign fibers 
from the background of cotton. 
3 EXPERIMENT OF SELECTION OF OPTIMAL 
OPERATING POINT 
Since the image features of foreign fibers are a function of 
wavelength and light energy, we set out to find the optimal 
wavelength and optimal light energy for these detected 
foreign fibers. That is to say, we need to establish several 
optimal operating points for the multiwavelength imaging 
system to detect these foreign fibers. 
The light source used in the multiwavelength imaging system 
was LED array, which emitted different wavelength in the 
visible and NIR region. Seven discrete bands, whose center 
wavelength at 405nm, 470nm, 520nm, 580nm, 630nm, 850nm, 
and 940nm were available. Also, the multispectral CCD was 
sufficient sensitive to the band from 400nm to 1 lOOnm. 
Six kinds of foreign fibers including hair, knitting, jute, 
plastic, wool, and bristle were chosen for this experiment. The 
choice of these foreign fibers over other types is justified 
because they cannot not only be distinguished visually from 
the background of cotton, but also they frequently and readily 
survive the ginning process and show up in the yam even the 
finished fabric. Sample of six kinds of foreign fibers were 
placed on the surface of cotton and then presented to the 
spectral imaging system at deferent wavelength. 
To evaluate the potential of each available wavelength and 
light energy for discriminating foreign fibers from cotton, the 
image feature discrimination between cotton and foreign 
fibers at different OOP was computed. 
In the spectral imaging experiment, light energy coefficient 
g — 
was defined as , where g represents the average 
gray level of an image. And feature discrimination was 
defined as the gray level difference between foreign fibers 
and cotton. Through the experiment, the OOP for each type of 
foreign fibers was quantitatively presented in table 1. 
Table.l indicated that different foreign fibers have the 
different OOP for detecting them over the spectrum (405nm 
to 940nm). Clearly, the band at 405nm could be an 
appropriate wavelength for detection of jute, knitting, wool, 
and bristle, but it is not effective to detect hair, plastic. 
Furthermore, these foreign fibers have the different light 
energy for detecting them despite the same detection 
wavelength. At the same time, the most distinctive different 
between hair, plastic and cotton occurred in the band at 
850nm. It is therefore clear that either the 405 or 850 nm 
wavelength as a center band can be used for the determination 
of foreign fibers on cotton in this study. For each specific type 
of foreign fibers, we can find the optimal wavelength and 
light energy to detect them. 
Foreign fibers 
Center wavelength Light energy 
coefficient 
hair 
850nm 
0.7053 
knitting 
405nm 
0.4316 
jute 
405nm 
0.7449 
plastic 
850nm 
0.7331 
wool 
405nm 
0.7831 
bristle 
405nm 
0.5367 
Table 1 the optimal operating point for detecting foreign 
fibers 
Based on the above results, two optimal bands for use in a 
multiwavelength detection scheme were selected. Table 1 
summarized the band at 405nm and 850nm were optimal 
wavelength for detecting the six types of foreign fibers. 
Although there are no single waveband images that can be 
used easily to differentiate foreign fibers from cotton, it is 
possible that the fused image of the two bands, which have