Paper summary 4

Convolutional Neural Network with data Augmentation for SAT Target Recognition

IEEE Geoscience and Remote Sensing Letters  IF:2.228

Target:  SAR data has some problems while use it to do the target recognition, such as lack of pose, target translation, random speckle noise. The purpose of this paper is using Data Augmentation knowledge  to revise and enlarge the data amount, after that, check by CNN.

Some tricks:

  1. Show the problems of SAR data
  2. The way of solving problems: different data augmentation methods
    1. Target translation: There are some difference between the training data and testing data, by convolution operation method(translation invariance)
    2. Speckle Noising: exponential distribution
    3. lack of pose: Pose synthesis such as rotate the image
  3. Use CNN and other Machine learning methods to detect the result

Conclusions:

  • This paper is not highly recommended, since the method is very simple, just use the data augmentation method as it has been mentioned in AlexNet paper
  • How to implement the Data Augmentation method in my research, such as
    • Rotate the image
    • Sub-sampling
    • Use PCA to get some strong feature in the training data

 

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