Parameters in ANN method using ENVI4.5 - NV5 Geospatial

Prev Go to previous topic

Go to next topic

Last Post 05 Jul 2009 03:51 AM by anon

Parameters in ANN method using ENVI4.5

3 Replies

Sort:


You are not authorized to post a reply.

Author

Messages

anon

New Member

Posts:

05 Jul 2009 03:51 AM
Dear users, I am conducting a research about image classification using ANN method included in ENVI 4.5. The ENVI4.5 offers users a friendly interface to fill out the parameters involved in the construction of an ANN. These parameters consist of activation, training threshold, training rate, momentum, number of hidden layers and number of training iterations etc. Unfortunately, there is no information about the number of neurons per hidden layer. Actually, the number of neurons is considerably important parameter and determination of it will significantly affect the performance of ANN for land cover classification. So I am wondering that how many neurons per hidden layer adopted in ENVI 4.5 (default), and whether users could change the number of neurons per layer for specific application, and how? Anyone has an idea? Any help would be appreciated! Best regards, peter.

Deleted User

New Member

Posts:

MariM

Veteran Member

Posts:2396
Veteran Member

21 Dec 2011 07:43 AM
Do you mean the number of nodes? I believe the number of nodes on the hidden layers is equal to the number of bands. The first input layer also has number-of-bands nodes and the output layer has number-of-classes nodes.

Deleted User

New Member

Posts:

27 Dec 2011 06:37 AM
[QUOTE]MariM wrote Do you mean the number of nodes? I believe the number of nodes on the hidden layers is equal to the number of bands. The first input layer also has number-of-bands nodes and the output layer has number-of-classes nodes.[/QUOTE] Marim, You are right with no. of input nodes and no. of output nodes.But no of nodes on the hidden layers is not always equal to number of bands.it can be more than that! I wish envi has the possibility to set up no. of nodes in hidden layer, i didnt have to turn to MATLAB !

You are not authorized to post a reply.