Information Technology Reference
In-Depth Information
such a happening, it is necessary to eradicate them before the invasive alien species
explosively breed.
Once breeding is starting, to clean the black bass of all is difficult. Therefore, the
efficient way to protect the ecological environment of native species is prevent the
growth and disinfect it in a early stages if the invasive alien species has invaded.
Proposed method intend to set up a camera in the water of ponds and lakes that do
not have the invasive alien species such as black bass still. If the alien species in-
vade, find them immediately. The approach that we think is first using AdaBoost to
identify whether there is a fish exist in the picture of a camera. If there is a fish exist,
then cut out the target area. Next, select several local area from the fish body, calcu-
late the value of several parameters based on co-correlation matrix texture analysis.
Then determine whether the target is a black bass or not.
In this study, we create a novel fish area detector for applying the AdaBoost. Then
use the co-occurrence matrix for texture analysis to three kinds of fishes, black bass,
carp and crucian, which are biological habitat same. From the evaluation value,
we can finding the available parameters which can separate the target species from
others by the body surface texture pattern.
In chapter 2, several target detection methods are introduced. In chapter 3, tex-
ture analysis methods are introduced, and the co-occurrence matrix method is se-
lected for this work. Simulations for fish detection with AdaBoost and specified
fish species identification are introduced in chapter 4 and finally chapter 5 is the
conclusion.
2
Fish Area Detection
2.1
Overview of Target Detection
For the subject detection technique, template matching is a traditional way. In the
template matching, a standard target pattern is manually predefined. Given an input
image, the cross correlation value with the standard pattern is calculated for each
local area separately. The existence of target is determined based on the correlation
values. This method has the advantage of easy implementation. However, it has
proven to be inadequate for creatures detection because it cannot effectively deal
with the variation in scale, pose and shape. Furthermore, a lot of computing resource
is need, the calculation speed is slow[1, 2, 3, 4].
Appearance-Based Methods is another target finding algorithm. Compare to
the template matching methods, the template in appearance method are learned
from the examples in the images but not predefined manually. Eigenfaces[5, 6,
7], Distribution-Based Methods[8, 9], Neural Networks[10, 11], Support Vector
Machines[12, 13], Naive Bayes Classifier[14, 15, 16], Hidden Markov Model[17]
are all developed for subject recognition or identification. But all of these approach
are time consumption, not suit for a real time application until the algorithm below
appeared.
Search WWH ::
Custom Search