Papers summary 1

good articles about Deep learning and its applications

1. Transferring Deep Convolutional Neural Networks for the Scene Classification of High-Resolution Remote sensing Imagery

It’s about the transfer learning method using in remote sensing field

Value: Transfer learning, using pretrained model, and using a few training samples to get good result.

target:

1. Investigate how to transfer features from these successfully pre-trained CNNs for HRRS scene classification with limited training data
2. Scenes classification

using many networks to try, including:

1. AlexNet 2.CaffeNet 3.VGG-F 4.VGG-M 5.VGG-S 6.VGG-VD16 7.VGG-VD19 8.PlacesNet

Dataset:

1. UCM
2. WHU-RS

 

Some tricks:

1. We usually only using the final layers, actually using middle layers is also effective
2. Based on the trained model
3. Only a few training data
4. By removing the last few layers of a CNN, treat the remainder of the CNNs as a fixed feature extractor
   1. regard the activation vector of the fully-connected layer as the global feature representations for scenes, using other sample classifiers such as linear SVM to be the classifier
   2. generate dense CNN activations from the last or penultimate convolutional layer with multiple scales, aggregate the obtained features into a global representation via the conventional feature coding scheme: such as BOW, Fisher encoding….
5. Compare the different feature coding methods, effect of different convolutional layers,  and compare with low-level features
6. Compared all the result, and proved then corresponding method is state-of -the-art

Others:

1. To intuitively understand the CNN activations, visualize the representation of each layers by inverting them into reconstruction images with the technique proposed in the paper called: Understanding Deep Image Representations by inverting them.
2. Alex,Caffe,VGG-F and PlacesNet have the similar structure.

 

2. Classification and Segmentation of Satellite Orthoimagery using convolutional neural networks

Best point is about using multiple CNN layers, and pretrained CNN filters with K-meaning

Target

1. Classify different landscapes, especially buildings

Some Tricks:

1. Using near infrared bands, totally 14 bands, some of them overlapped
2. Digital surface model(DSM), increases classification accuracy by providing height information that can help distinguish between similar looking categories.
3. Using a GUI to manually labeling, pre-processing using the cluster algorithm called simple linear iterative clustering
4. pixel to be classified is located at the center
5. filter of CNN pretrained by unsupervised methods such as K-meaning
6. Using multiple CNNs, concatenated as in put for the fully-connected layer. Two different methods
   1. Different types of CNN structure between each level
   2. Different input, same CNN structure
7. Post-Processing
   1. Classification Average: can get rid of salt-and-pepper
   2. Region merging: merge the result with original segmentation result, set a threshold, two adjacent segments were merged if the average classification accuracy of all the pixels within each  segment were over the threshold
8. Prove and compare many aspacts
   1. how to choose the best architecture: Window size, filter size, filter number, pooling dimension
   2. best number of CNN
   3. Influence of the number of CNN in parallel
      1. different CNN numbers
      2. different parameter settings
9. Good result come out when combine:
   1. aforementioned best trucure
   2. Average combine
   3. Region Merging

Others:

1. Very good generalization of CNN using in Remote sensing