23.04.2016 Views

DEEP LEARNING APPROACH FOR REMOTE SENSING IMAGE ANALYSIS

D2_1710_03_Benoit

D2_1710_03_Benoit

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Results :<br />

Accuracy vs training dataset size<br />

Accuracy<br />

102<br />

Accuracy on Pavia University dataset<br />

100<br />

98<br />

96<br />

94<br />

92<br />

90<br />

6 layers, 3*3 neighbors, ~4419 parameters<br />

6 layers, 5*5 neighbors, ~6074 parameters<br />

88<br />

0 10 20 30 40 50 60 70 80 90 100<br />

Training samples ratio (%)<br />

CNN challenger, 5*5<br />

neighbors, no pretraining<br />

K. Makantasis&al “Deep<br />

supervised learning for<br />

hyperspectral data<br />

classification through<br />

convolutional neural<br />

networks,” IGRS2015<br />

~20000 parameters<br />

SAE challenger, 7*7<br />

neighbors, with pretraining<br />

X. Ma&al“Hyperspectral image<br />

classification via<br />

contextual deep learning,”<br />

EURASIP JIVP 2015<br />

>>20000 parameters

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!