Identify a Specified Fish Species by the Co-occurrence Matrix and AdaBoost - Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing - page 102

()

()

() () () ()

Fig. 7 Body texture pattern of crucian

Finally, there are things to note about the value of the feature be calculated. For
example, the formula (5) ASM is used to measure whether a distribution which
elements of the co-occurrence matrix is to position around. Thus, the uniformity of
the texture can be measured. However, when an element of co-occurrence matrix
are concentrated in the vicinity other than the diagonal, texture is not uniform.
As mentioned above, in some case it is unknown what you are physically mea-
suring, it is necessary to pay attention to the value of the feature quantity calculated.
Feature amount of the other is the same.

4 Simulations

By using the proposed method, an experiment was carried out to verify whether it
is possible to identify the black bass from other fish species.

4.1 Fish Recognition Rate Experiment

In this experiment, using the Haar-Like as the image feature that is implemented in
OpenCV, and used the AdaBoost as the learning algorithm, we produced an original
fish detector. Then using the previously created fish detector to determine the error
recognition rate and recognition rate. Because we have not a practical environment
can be used, this experiment is performed under a ideal condition, The result is
shown in Fig. 2. Although there was no error happened, it is necessary to enhance
the study file to improve the availability in a practical environment.

4.2 Body Surface Texture Pattern Analysis

In this experiment, three kinds of freshwater fishes(black bass, carp, crucian) body
texture are quantified by using the co-occurrence matrix, and then verify whether it
is possible to identify the black bass from the two others. This time, using the lateral