Feature fusion methods in remote sensing

Today, I learned some feature fusion methods as follows (refer to the literature of qinrongjun 3D CD reviews):

1. Post-refinement

2. Direct feature fusion: considering all channels of information simultaneously

3. Post-classification comparison: first, conducting objects extration or classification by changeing the spectral values or feature values into label images. Then, compare labels.

 

The considering part is part 2.

1) SOM: self-organizing map (a kind of clustering methods like k-means)  it considers not only its own cluster, but also has an influence on neighbouring clusters. ( code: refer to MATLAB)

2) MNF: minimum noise fraction which improves PCA methods by replacing the variance-based order with noise-based one, with an increase of noise robustness.

 

date: 20190225