MATLAB coding for artificial neural network (ANN) to automatically determine the species of Gal´apagos finches (birds of the subfamily Geospizinae) in images

Prof. S. wants to use an artificial neural network (ANN) to automatically determine the species of Gal´apagos finches (birds of the subfamily Geospizinae) in images using the following measurements: Beak length, beak height, eye diameter, head length, and body length. Given the location where the pictures were taken, the possible species are: 1. Large Ground Finch Geospiza magnirostris, 2. Medium Ground Finch Geospiza fortis, 3. Small Tree Finch Camarhynchus (formerly Geospiza) parvulus, and 4. Green Warbler-Finch Certhidea olivacea. He has a database of a few hundred labeled images of individuals of these species on which to train his ANN.

Provide your input on the following aspects for this ANN. That is, give choices for these aspects and justify your choices. Prepare a technical report on your proposed network which including the elements shown as follows:

A. What learning mechanism(s) should the ANN use?

B. How many input units should the ANN have?

C. How many output units should the ANN have?

D. Should the ANN use hidden units or not?

E. Should the ANN use feedforward connections, recurrent connections, both, or neither?

F. What activation function(s) should the neurons use?

Answer 1

[Krishna]

Hello Ramos,

I have provided with possible answers to each question, please have a look.

• Learning mechanism that ANN uses is Backpropagation.
• Input units of ANN should be equal to the number of features. In your case it would be 5.
• Output units depend on the number of classes you want to classify and whether you want to use label encoding or one hot encoding.
• Depend on the complexity of the problem. You can try and check how many hidden layers and number of neurons work the best.
• For simple classification feed forward network are best. RNN are not used for classification problems.
• You should use SoftMax activation function for the output layer and ReLU or tanh activation function for the hidden layer.