Paper summary 5

Deep learning earth observation classification using imageNet pretrained Networks

IEEE Geoscience and remote sensing letters, IF: 2.228

Target: Science classification

Some trick:

  1. use pretrained model, reason:
    • using CNNs along with limited labeled data can be problematic
    • Actually the data set can not seem as big data set
  2. such kind of method still not be used (actually not)
  3. use different training set, and set it as so-called transfer learning
  4. show the scatterplots graph

High-Resolution SAR Image Classification via Deep Convolutional Autoencoders

IEEE Geoscience and remote sensing letters, IF: 2.228

Target: land feature pixel-based classification using SAR image

Some tricks:

  1. mentioned the methods of feature extraction:
    1. statistical method
    2. transform domain method
    3. model based method
  2. SAR data, not RGB
  3. The former method lack the ability of extracting feature, therefore using CNN
  4. Use a so-called structure stacked autoencoder(SAE), just like neural networks
  5. mentioned the contribution
    1. DCAE(deep convolutional autoencoder) is proposed to extract features automatically
    2. scale transformation (actually this is also CNN’s characteristic)
  6. Postprocessing
    1. use some morphological smoothing method such as: region detection, dilation and erosion

Leave a Reply

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

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

Google+ photo

You are commenting using your Google+ 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 )


Connecting to %s