Abstract:
Sexing is a difficult task for most birds (especially ornamental birds) involving
expensive, state
-
of
-
the
-
art equipm
ent and experiments. An intelligent fowl sexing system
was developed based on data mining methods to distinguish hen from cock hatchlings.
The vocalization of one
-
day
-
old hatchlings was captured by a microphone and a sound
card. To obtain more accurate inf
ormation from the recordings, time
-
domain sound
signals were converted into the frequency domain and the time
-
frequency domain using
Fourier transform and discrete wavelet transform, respectively. During data
-
mining from
signals of these three domains, 25
statistical features were extracted. The Improved
Distance Evaluation (IDE) method was used to select the best features and also to reduce
the classifier's input dimensions. Fowls’ sound signals were classified by Support Vector
Machine (SVM) with a Gaussi
an Radial Basis Function (GRBF). This classifier identified
and classified cocks and hens based on the selected features from time, frequency and
time
-
frequency domains. The highest accuracy of the SVM at time, frequency and time
-
frequency domains was 68.5
1, 70.37 and 90.74 percent, respectively. Results showed that
the proposed system can successfully distinguish between Hen and Cock hatchlings. The
results further suggest that signal processing and feature selection methods can maximize
the classification
accuracy.
Keywords:
Gender de
termination,
Non
-
invasive sexing, Animals behavior, Fowls
vocalization,
Signals processing.
