Re: How to write a decision in ERDAS?

33 views
Skip to first unread message

ERDAS Communities - Automated Email

unread,
Nov 9, 2010, 12:05:51 PM11/9/10
to erdas-communitie...@googlegroups.com

Re: How to write a decision in ERDAS?
By Timo Ikola in ERDAS Software Forum

As always Donn's comment hits the target. Perhaps conditional statement in Modeller is the trick that makes it easiest. My comment about Knowledge engineer was more like a one possible way rather than excat instruction.  I have not used it so much that I can immediately say how to do it in it.

 

Knowledge engineer is a rule based classification tool which in practise is a big tree of conditional statements. Benefit against model maker is that it is more flexible and modern for example when calling external applications or other information sources.

 

timo

----------------------------------
You were sent this email because you opted to receive email notifications when someone responded to this thread.

To unsubscribe:

  1. To unsubscribe from this item only, visit the item link and turn off notifications.
  2. To stop all notifications of any updated items from ERDAS Communities, go to Edit Profile, click on the Site Options tab, then under Email Configuration, set Send Notifications to "No".

ERDAS Communities - Automated Email

unread,
Nov 9, 2010, 12:45:54 PM11/9/10
to erdas-communitie...@googlegroups.com

By Zhuoting in ERDAS Software Forum

Donn and Timo,

Thanks so much for your replies. I think I probably will go with the Modeler for now. My tree will grow bigger as I tried to do a better job classifying, and I hope this will not get too tricky.

Thanks a lot!

Zhuoting

ERDAS Communities - Automated Email

unread,
Nov 10, 2010, 4:06:35 PM11/10/10
to erdas-communitie...@googlegroups.com

Re: How to write a decision in ERDAS?
By Zhuoting in ERDAS Software Forum

I'm not sure the modeler will work. I brought in some new data such as DEM and precipitation to do a better job classifying the image, and I ended up with a 200-node tree in ENVI. I tried to write that tree in modeler, and I'm totally exhausted just keeping track of the nodes. I really hope there is a better way to do this, such as the Knowledge Engineer, and I don't know actually how to use it?

I'll very appreciate any help. Thanks!

Zhuoting

ERDAS Communities - Automated Email

unread,
Nov 10, 2010, 5:36:39 PM11/10/10
to erdas-communitie...@googlegroups.com

By Donn Rodekohr in ERDAS Software Forum

Zhuoting,

I apologize for my ignorance, but I do not know the logic of the ENVI tree.  If you could explain what the ENVI node tree is and how it is created perhaps we can determine what tools in ERDAS could be used to accomplish the task.  I suspect that you are correct that writing a CONDITIONAL statement to handle 200+ conditions would be excessive.  Perhaps Knowledge Engineer is the tool, perhaps Objectives, perhaps a spatial model.  

Until I have an idea about how ENVI nodes work, I cannot come up with an acceptable work plan.

ERDAS Communities - Automated Email

unread,
Nov 10, 2010, 10:52:56 PM11/10/10
to erdas-communitie...@googlegroups.com

By Zhuoting in ERDAS Software Forum

Donn,

I should have explained this a little bit more, thanks for bringing it up. What I did in ENVI was selecting training pixels for all three classes (irrigated, rainfed, and others), and I used the training pixels to develop a decision tree. It's really easy to show the logic of the ENVI tree using a jpg file, but I don't see an option attaching a picture here. Basically, I used 16 bands to separate out three classes, for example, in ENVI tree:

Node1: B8 <= 169.4, if yes, goes to Node2: B1 <= 128.9, if yes, goes to Node 4: B12 <= 147.0, if yes, goes to Node6: B4 <= 167.0, if yes, goes to Node8: B2 <= 128.5, if yes, goes to Node12: Rainfed, and this is a terminal node.

I have over 100 intermediate nodes, and 63 terminal nodes, so it means different combination of bands value yes/no will lead to irrigated / rainfed / others classes. If I use Conditional, I need to write over 100 Conditionals, and I have to keep track yes/no paths through the tree.

I hope this makes sense to you, if you have an email address, I can send a jpg file to you to show the structure of the tree. I was just wondering if there may be something other than the modeler that can make my life easier?

Thanks a lot!

Zhuoting

 

 

 

 

ERDAS Communities - Automated Email

unread,
Nov 11, 2010, 3:03:16 AM11/11/10
to erdas-communitie...@googlegroups.com

By Arto Vuorela in ERDAS Software Forum

From the Options tab it is possible to enter an attachment.

ERDAS Communities - Automated Email

unread,
Nov 11, 2010, 5:30:41 PM11/11/10
to erdas-communitie...@googlegroups.com

By Timo Ikola in ERDAS Software Forum

As a graphical this looks much like what Erdas Knowledge Engineer does. which then is more handy ENVI or Erdas is probably quite much upon what tool user has more used.

 

I still think that Knowledge Engineer would be right tool in Erdas world. It is often underestimated and things are done in modeller but I agree that handling all those descions in modeller equation box might get tricky when amount of equation increases. So learning Knowledge Engineer here might tur case from text to graphics.

 

timo

ERDAS Communities - Automated Email

unread,
Nov 11, 2010, 5:40:53 PM11/11/10
to erdas-communitie...@googlegroups.com

Re: How to write a decision in ERDAS?
By Timo Ikola in ERDAS Software Forum

Knowledge engineer forms out from three objects,

Hypothesis (green box) can represent the output classes

Rules (yellow box) represents the conditions that must be filled so that rule is true or false and by then defines the hypothesis correctnes

Variables (blue boxes) are the information that is taken as a input to rule.

 

So variable goes to rule. In rule true or false is defined by values in variables and after rule is true it can form a hypothesis that is the output class.

In professional level tour guide there is document about Knowledge classifier and that should lead you in beginning. You can find this doc from 2010 help menu.

Is this handy and suitable tool for your need is something you must evaluate by yourself but in my opinion it might be one suitable option as a addition to Model Maker conditional stetements.

 

timo

ERDAS Communities - Automated Email

unread,
Nov 11, 2010, 5:50:57 PM11/11/10
to erdas-communitie...@googlegroups.com

By Zhuoting in ERDAS Software Forum

Hi Timo,

Thanks so much for your responses. I think Knowledge engineer should be the way to go. It'll be a new learning experience for me. Thanks!

Zhuoting

ERDAS Communities - Automated Email

unread,
Nov 12, 2010, 9:27:07 AM11/12/10
to erdas-communitie...@googlegroups.com

By Donn Rodekohr in ERDAS Software Forum

After looking at the tree diagram I totally agree with Timo.  

Given enough time and knowing that the frustration level could be quite high, a Spatial Modeler solution could be worked out.  But Knowledge Engineer is better designed to handle this scenario.

ERDAS Communities - Automated Email

unread,
Nov 12, 2010, 10:32:11 AM11/12/10
to erdas-communitie...@googlegroups.com

By Timo Ikola in ERDAS Software Forum

Knowledge engineer is often underestimated tool - it is actually fairly handy when cleaning up classified thematic data by rules. For excample dark pixel is not water if there is strong slope type thing. I have used it for that and it works nicely.

Model maker is also good and in some tasks these tools overlap a bit.

One strenght of Knowledge Engineer is that you can use external exe, vector data etc. in bit more handy way there.

 

timo

Reply all
Reply to author
Forward
0 new messages