Tile-Based Semisupervised Classification of Large-Scale VHR Remote Sensing Images

Table 1

Parameters used by various feature extraction techniques.

Feature type	Feature vector length	Note
HOG	4365
LBP	256	8 neighbourhood, uniform pattern
BOW	4000	Vocabulary size = 4000
CNN—vgg-very-deep-16	4096
CNN—GoogLeNet	1024
CNN—vgg-m	1024
CNN—ResNet-152	2048