3.2.4. Other algorithms . . . . . . . .................................................................................. 64
3.3. Classification . . . . ................................................................................................ 65
3.3.1. Fuzzy clustering and neural network . . . . . . . .................................................................. 65
3.3.2. Rough sets. . . . . . . . . . . . . .................................................................................. 65
3.3.3. Support vector machines . .................................................................................. 65
4. Applications and case studies . . .......................................................................................... 65
4.1. On-line foreign fibers inspection with an AVI system . . . . . . ............................................................. 65
4.2. Multi-wavelength imaging system. . . . . . ............................................................................. 66
4.3. White light and UV light alternating system. . . . . . . . . . . . . . ............................................................. 66
4.4. X-ray imaging system . . . . . . . . . . . . . . . . ............................................................................. 66
4.5. Linear laser imaging system . . . . . . . . . . . ............................................................................. 67
4.6. Hyperspectral imaging system . . . . . . . . . ............................................................................. 67
5. Conclusions and future perspectives. . . . . . . . . . . . . . . . ....................................................................... 68
Acknowledgments . . . . . . . . . . . .......................................................................................... 68
References . .......................................................................................................... 68
1. Introduction
Cotton is an important resource related to numerous nations’
economies. Cotton processing and cotton spinning play an impor-
tant role in the Chinese national economy in particular. When har-
vesting, cotton is subject to contamination from numerous sources,
and a variety of foreign matter may become mixed with raw cotton
during cotton processing. Foreign matter, also called foreign mate-
rials, foreign contaminants, or cotton trash, refers to both botanical
trash and non-botanical trash which are inadvertently mixed with
cotton during picking, storing, drying, transporting, purchasing,
and processing (Yang et al., 2009a). Botanical trash includes hull,
leaf, bark, seed coat, etc. Non-botanical trash is also called foreign
fibers including hair, binding rope, plastic film, candy wrappers,
polypropylene twine, etc. Foreign matter is difficult to remove
and is easily broken into countless tiny parts that in turn increase
the breakability of cotton yarn and reduce the processing efficiency
and market value according to the cotton grading system. Foreign
matter also affects the quality of yarn and woven cloth as well as
the appearance of dyed cloth.
Conventional detection methods for foreign matter in cotton
have been performed by human workers, but most of these manual
inspections are time-consuming, inefficient, and have unverifiable
accuracy rates (Yang et al., 2009b; Li et al., 2006). The poor perfor-
mance of conventional detection methods and the bad effect of for-
eign matter on cotton industry have attracted great attention from
research institutes and cotton enterprises (Luo, 2007). Researchers
in China and abroad have been conducting research on foreign
matter detection for some time and have made significant pro-
gress. Several instrumental and sensory methods have been devel-
oped for the detection of foreign matter. The main detecting
principles can be categorized into three types: photoelectric detec-
tion, ultrasonic detection, and optical recognition (Shi, 2007).
Photoelectric detection detects foreign matter in cotton using a
phototransistor. It is used to identify foreign matter according to
the color variation between cotton and foreign matter. The method
is simple and has a low manufacturing cost. However, the detec-
tion rate is low due to the great attenuation of sensitivity and poor
stability of the phototransistor. It is also impossible to identify tiny
colored foreign matter as well as foreign matter that has a similar
brightness to cotton, such as white polypropylene (Chang, 2006).
Ultrasonic sensors transmit ultrasonic waves at the cotton and
receive the reflected information. Ultrasonic signals reflected by
objects with different densities are different. Therefore, the foreign
matter is identified by signal processing and comparison because
the signals reflected by foreign matter are typically stronger than
those reflected by cotton. It can be used to detect certain types of
foreign matter such as bulked paper strips, cloth strips, plastics,
etc. However, the speed of ultrasonic transmission is lower than
that of light, so the identification process is slow and sometimes
cannot be completed in the time allotted. In addition, ultrasonic
sensors are not able to identify small foreign matter (Chang, 2006).
Using a high-speed CCD or CMOS camera, the optical detection
method scans the surface of the cotton layer, and the images gen-
erated from the scanning signals are sent to a computer system for
processing. Both line scan cameras and area scan cameras can be
used, the first of which is more flexible and convenient. In contrast
to the two former methods, the advantage to optical detection is
that it can recognize small foreign matter in cotton and meet the
requirements of real time inspection. The only disadvantage is
the high cost of manufacturing the system (Chang, 2006). However,
with scientific and technological development, the cost will drop.
Based on the optical detection principle, computer vision tech-
niques have the advantages of cost effectiveness, consistency,
superior speed, objectiveness, and accuracy. With the advances
in hardware and software for digital image processing, automatic
inspection systems known as computer vision or machine vision,
mainly based on camera-computer technology, have been investi-
gated for the sensory analysis of agricultural and food products and
have been proven successful for the objective measurement of var-
ious agricultural products (Brosnan and Sun, 2004). Furthermore,
applications of these techniques have now expanded to various
areas such as medical diagnosis, automatic manufacturing and sur-
veillance, remote sensing, technical diagnostics, and autonomous
vehicle and robot guidance (Brosnan and Sun, 2002). In recent
years, computer vision systems have been applied to the textile
industries (Tantaswadi et al., 1999; Millman et al., 2001;
Abouelela et al., 2005) for inspection and/or removal of foreign
matter in cotton (Lieberman et al., 1998) and wool (Zhang et al.,
2005a–c; Su et al., 2006). These systems hold great potential for
the inspection of cotton foreign matter.
Fig. 1. Components of a basic computer vision system.
60 H. Zhang, D. Li / Computers and Electronics in Agriculture 109 (2014) 59–70