Paper summary 2

Convolutional neural network based automatic object detection on aerial images

Letter: IEEE GRSL IF: 2.228

Target: Scene classification

Some tricks:

  1. Using the model generated from one dataset to test another dataset.
  2. using pretrained VGG model by Caffe framework, except the final layer, because the classification target amount is not equal, to avoid overfitting. Just like the project we do in Kaggle, directly import trained VGG model. Its not about transfer learning methods
  3. Using Adagrad rather than SGD. Adagrad is an algorithm for gradient-based optimization that does just this: It adapts the learning rate to the parameters, performing larger updates for infrequent and smaller updates for frequent parameters. For this reason, it is well-suited for dealing with sparse data.
  4. based on cell, not on pixel
  5. the future work will also focus on multiple layers

Others:

  1. Mentioned that using a pretrained network proved to have better performance
  2. multiscale images input can improve the representation of size-varying objects
  3. often using UCM as a classification evaluation

 

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s