Geography Reference
In-Depth Information
could not be correctly fitted (e.g. neighbouring fish and
fish whose orientation was very different from the initial
PDM). Zion et al. (1999) extracted typical features from
dead fish tails and used them for species identification.
For three species (Common carp,
Cyprinus carpio
; St.
Peter's fish,
Oreochromis
sp. and grey mullet,
Mugil
cephalus
), the average identification accuracy was higher
than 93%. More recently, Zion et al. (2007) improved
their image-processing algorithms by extracting size-
and orientation features from the fish silhouettes
(Figure 17.5). The overall species recognition accuracy
(from swimming fish) was about 98%.
Recent developments have improved the automated
species recognition for species that have similar shape
characters (e.g. seven salmonid species with similar mor-
phology: Lee et al., 2004) or IA from low-quality images
(i.e. images lacking distinctive or stable morphological
features: Rova et al., 2007; detection of target species in
turbid habitats from dual-frequency sonar techniques:
Frias-Torres and Luo, 2009). White et al. (2006) devel-
oped a 'computer vision machine' for identifying and
measuring different species of fish from 10 shape and
114 colour features. The fish were transported along a
conveyor belt underneath a digital camera. The image
processing algorithms determined the orientation of the
fish, identified flat or round body shapes (with 100%
accuracy), measured fish length (with a standard devia-
tion of 1.2 mm) and differentiated between seven species
(with up to 99.8% sorting reliability). This machine could
theoretically process up to 30,000 fish per hour using a
single conveyor belt based system.
Body shape descriptors are less sensitive to lighting
variations and water quality and are generally preferred
to colour features for species recognition. However, in
some cases, colour restoration of underwater images
can be achieved (Iqbal et al., 2007; Figure 17.6) and
Figure 17.6
Example of an underwater image before (left) and
after (right) enhancement. Reproduced from Iqbal, K. et al.
(2007) Underwater image enhancement using an integrated
colour model. IAENG International Journal of Computer
Science, 34, 2.
©
Copyright International Association of
Engineers.
the subsequent restored images should give better
results when displayed or processed (see an example of
fish segmentation and feature extraction in Chambah
et al., 2004).
Fish species recognition can also be performed by
studying the shape of anatomical structures such as
otoliths.
3
The general morphology of the saccular otoliths
(i.e. the largest pair among three in teleosts) is usually
species specific and has been used for species identifica-
tion using IA (e.g. Tuset et al., 2003). Fourier analysis has
traditionally been used to study otolith morphology, since
it is an effective method for describing outline shapes, but
it does not encourage intuitive understanding of the rea-
son for subtle shape differences (Cadrin and Friedland,
1999). Studying numerous otolith shape descriptors (e.g.
aspect ratio, compactness, eccentricity, ellipticity, bilateral
symmetry), Tuset et al. (2006) demonstrated that species
differentiation from otolith shape characterisation should
be enhanced by standardising variables with respect to
fish length and adding otolith weight in the analysis.
17.2.1.2 Stock differentiation
Variability in growth, development, and maturation
creates a variety of body shapes within a species, and iden-
tifying discrete units of the stock is a basic requirement for
fisheries science and management. Geographic variation
in morphometry has been used to discriminate 'pheno-
typic stocks' of fish (defined as groups with similar growth,
mortality, and reproductive rates, Cadrin, 2000) for over
130 years (Heinke, 1878 cited by Cadrin, 2000). For
example, Corti et al. (1988) detected variations between
six strains of common carp (
Cyprinus carpio
) from truss
Figure 17.5
Left: a carp image acquired by a real-time
underwater system, in a laboratory pool. The background light
is reflected from the apparatus frame. Right: segmented contour
and landmarks found by the system. Reprinted from
Computers and Electronics in Agriculture, 56, Zion et al.,
Real-time underwater sorting of edible fish species, 34-45,
Copyright 2007, with permission from Elsevier.
3
Otoliths are aragonitic mineralisations positioned in the mem-
braneous labyrinth of the inner ear of bony fishes and play an
important role in the senses of hearing and balance (see Popper
et al., 2005 for recent review).
Search WWH ::
Custom Search