Table 8 Effect of different deep learning models on the evaluation criteria values for Quick Bird images.
From: Remote sensing image description based on word embedding and end-to-end deep learning
Model features | Fusion | Word | |||||
|---|---|---|---|---|---|---|---|
P | R | F | P | R | F | ||
CNN 25 | C | 0.8732 | 0.8430 | 0.8593 | 0.8809 | 0.8471 | 0.8632 |
Bi-LSTM28 | B-L | 0.8647 | 0.8388 | 0.8516 | 0.8799 | 0.8452 | 0.8517 |
Attention 29 | A | 0.8707 | 0.8399 | 0.8550 | 0.8801 | 0.8465 | 0.8629 |
Dense Net 31 | D-N | 0.8788 | 0.8451 | 0.8666 | 0.8822 | 0.8496 | 0.8656 |
Attention-CNN-Bi-LSTM | A-C-B | 0.8820 | 0.8483 | 0.8698 | 0.8926 | 0.8531 | 0.8724 |
Attention-CNN-IndRNN | A-C-I | 0.8811 | 0.8456 | 0.8630 | 0.8919 | 0.8524 | 0.8707 |
CNN_LSTM | C-L | 0.8761 | 0.8439 | 0.8597 | 0.8872 | 0.8503 | 0.8684 |
CNN-Bi-LSTM | C-B | 0.8817 | 0.8479 | 0.8645 | 0.8891 | 0.8510 | 0.8696 |
CNN-IndRNN | C-I | 0.8752 | 0.8446 | 0.8596 | 0.8835 | 0.8502 | 0.8665 |
End-to-end | E2E | 0.8905 | 0.8593 | 0.8778 | 0.9057 | 0.8741 | 0.8896 |
