How to make Neural Network Ignore the background?

6 ビュー (過去 30 日間)
Steven Moran
Steven Moran 2017 年 9 月 7 日
編集済み: Steven Moran 2017 年 9 月 13 日
I'm running deep learning on rocks, the trey the rocks sit on needs to be ignored, so I paint it a different color (neon green) and used a color thresholder generator to remove the background.
The problem is the 'removed' section still shows up as pixels on the image. So even when I 'remove' the background, the image itself still has all that empty space. I want the neural network to completely ignore it and only see the rocks.
Please help, for a really thorough solution I'll send you some bitcoin :)
  6 件のコメント
Image Analyst
Image Analyst 2017 年 9 月 13 日
An image must remain rectangular so you can't really "remove" the pixels. You can color them a different color like blue or green or black, but they are still there if you want it to be a 2-D image, as opposed to just a 1-D list of color values. Please define exactly what "remove" means to you. Do you just mean "ignore"? Or do you think (incorrectly) that the "tray" pixels can somehow be thrown away so they do not even appear in the image at all?
Steven Moran
Steven Moran 2017 年 9 月 13 日
I want the neural network to ignore the background.
Apparently that simply means I need to train a 'class' as the background itself, since the background can't be 'removed' fully?
I have 6 classes of rock types. Apparently I need to make the background Tray as a class also?

サインインしてコメントする。

回答 (2 件)

Greg Heath
Greg Heath 2017 年 9 月 13 日
You don't need to remove the background.
Just consider it another class.
Hope this helps.
Thank you for formally accepting my answer
Greg
  1 件のコメント
Steven Moran
Steven Moran 2017 年 9 月 13 日
Ok I'll try it this weekend. I'll let you know how it goes.

サインインしてコメントする。


Helper
Helper 2017 年 9 月 13 日
If you want to classify different types of rock, then you do not have to remove the background.
One of the powerful place of Neural Network is it could extract useful feature automatically (namely, remove the useless information automatically). Generally, the first several layers of Neural Network work as feature extractor, and the following layers work as a classifier.
  1 件のコメント
Steven Moran
Steven Moran 2017 年 9 月 13 日
編集済み: Steven Moran 2017 年 9 月 13 日
one of my rock classes is very smooth and dark(like the background trey surface). that's why Im thinking I need to paint the background and set the background as its own class. I'm also splitting the screen into 300x300 pixel segments so get the exact percentage of each rock type present in the sample. When I segment it into smaller portions, it thinks the background is a smooth rock. That's why I need to paint it and set it as its own class I think.

サインインしてコメントする。

カテゴリ

Help Center および File ExchangeImage Data Workflows についてさらに検索

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by