In this paper, we proposed a single neural network as a multi-class
pattern classifier to identify seven different species of butterflies. The
feature extracted from an image is the most important factor in the
success of automatic species identification systems based on images
of organisms. We used BLS entropy profiles (an assembly of BLS
entropies) as the feature for the proposed species identification
system based on an artificial neural network. All specimen images
were transformed to binary images, after which the boundary shapes
of the left wing were segregated. We selected 360 evenly distributed
points along the boundary of the shape, and calculated the BLS
entropy for each selected point.
We compared the accuracy of the network trained over BLS
entropy profiles with that of the network trained over binary vectors
to verify the effect of the former as a feature. This showed that a single
neural network system accompanied by BLS entropies calculated for
the pixels along the boundary achieved a high level of accuracy (over
86%) for the identification of butterfly species. In addition, training
the system over the training data set required only a few minutes.