X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 24 Jan 2011 11:13 AM by  anon
problemas con el SVM
 3 Replies
Sort:
You are not authorized to post a reply.
Author Messages

anon



New Member


Posts:
New Member


--
24 Jan 2011 11:13 AM
    Hola! estoy comenzando a trabajar con ENVI. En estos momentos estoy intentando realizar una clasificacion supervisada con SVM en un mosaico constituido por 4 imagenes y 10 clases. Sin embargo, luego de dejar que el programa corra durante toda una noche ocurren differentes errores (y he utilizado maquinas con distintas capacidades). Alguien ha tenido este problema? sabe como resolverlo? Muchisimas gracias! L.

    Deleted User



    New Member


    Posts:
    New Member


    --
    24 Jan 2011 03:09 PM
    I hope it's OK if I answer in English. I don't speak Spanish, but Google Translate did a pretty good translation of your question, so I think I understand the situation. With the information you've provided so far, it's hard to say what might be going wrong. Here are the questions that I have at this point: When you say that different errors occur, do you mean that you get multiple error messages while the classification is running? Or do you get different error messages each time you try to run the classification, when you choose exactly the same classification input data and parameters? Can you provide the error messages that you see? If you get more than one, it would be helpful to see all of the error messages that you have received. It sounds like your input image, which is a mosaic of 4 smaller images, is pretty big, if the SVM runs overnight. Exactly how big is it? How many columns and rows, and how many bands? What version of ENVI are you using? Do any of your ROIs have identical names? I see a case where that led to the SVM classification failing in earlier versions of ENVI. Did you set the Pyramid Levels parameter to the SVM classification to anything other than the default value? The answers to those questions may help us track down (or rule out) what is going wrong for you. - Peg

    Deleted User



    New Member


    Posts:
    New Member


    --
    26 Jan 2011 12:13 PM
    Hi Peg, Thank you very much for your help. I will try to write in english..and asnwer your questions. My mosaic have 4 image of Landsat TM 5 this is the caracteristic Dims: 11754 line x 22821 column x7 band Size: [Byte] 1,877,666,238 bytes. Pixel : 30 Meters Wavelength : None I use ENVI 4.7 version. I not use the same name for the ROIDs and today I use 11 different class. The errors... Well, I have try in various ways, with different computer and with values of pyramid level by defaul and by different values (2 and 3). Most common that happen is that the machine put on stand by (stop working) and the ENVI not responde and I must put in off the computer. When I use a notebook the error was: "unable to allocate memory : to make array the result maybe invalid. When I tried to use a portion of the mosaic (big portion, because with the small I have very good clasifications) (with subset of the image) the error was : "loop limit expresion too large for loop variable type"long (46364) the result maybe invalid. Other funtion that I would use was TERCAT (with defaul values) but when classified with SVM....the computer stop to working or another error with result invalid occur.... I never have problem with other classification but with the support vector machine is imposible class my mosaic..the problem is that with SVM my classification are good (but in small size). Thank you for your helps Best regards, L.A

    Deleted User



    New Member


    Posts:
    New Member


    --
    07 Feb 2011 01:35 PM
    That is indeed a very large image! I see why it was taking so long for the SVM classification to run. I wonder if when the machine is put on standby, what is happening is that it is just taking a very long time for the SVM classification? In other words, if it were possible to wait longer, maybe SVM would have finished? The unable to allocate memory to make array message means that at some point in the processing, ENVI needed to store more data in RAM than was available on your machine. That is pretty rare in ENVI, because it uses tiling to minimize the amount of RAM that it needs to allocate at any one time. But if you are running the SVM classification many time in a row, you may be running into a bug (CR 51276), caused by a memory leak in the SVM routine. The way to avoid this problem, for now, is to start a fresh ENVI session before you try running the SVM classification with this image. A complete fix for that bug has not yet been found. If that's not the issue (you are not running SVM multiple times before you get this error), then there must be something that SVM does internally that requires a very large array to be allocated in RAM, when working on very large images. In this case, you will want to contact ITT VIS Technical Support (support@ittvis.com or (303) 413-3920) or your local ENVI distributor to track down what is triggering this problem with your data. For all of the other behaviors you report, you will need to work with tech support to figure out what is going wrong. They may need to get a copy of your data, ROIs, and classification parameters to reproduce the problem and log a bug report. - Peg
    You are not authorized to post a reply.