HANet: A Hierarchical Attention Network for Change Detection With Bitemporal Very-High-Resolution Remote Sensing Images
Benefiting from the developments in deep learning technology, deep-learning-based algorithms employing automatic feature extraction have achieved remarkable performance on the change detection (CD) task. However, the performance of existing deep-learning-based CD methods is hindered by the imbalance between changed and unchanged pixels. To tackle this problem, a progressive foreground-balanced sampling strategy on the basis of not adding change information is proposed in this article to help the model accurately learn the features of the changed pixels during the early training process and thereby improve detection performance.Furthermore, we design a discriminative Siamese network, hierarchical attention network (HANet), which can integrate multiscale features and refine detailed features. The main part of HANet is the HAN module, which is a lightweight and effective self-attention mechanism. Extensive experiments and ablation studies on two CDdatasets with extremely unbalanced labels validate the effectiveness and efficiency of the proposed method.
PDF AbstractCode
Spaces
Tasks
S2Looking
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Uses Extra Training Data |
Benchmark |
|---|---|---|---|---|---|---|---|
| Change Detection | CDD Dataset (season-varying) | HANet | F1-Score | 89.23 | # 16 | ||
| Precision | 92.86 | # 5 | |||||
| F1 | 89.23 | # 5 | |||||
| Overall Accuracy | 97.32 | # 6 | |||||
| IoU | 90.55 | # 4 | |||||
| Recall | 85.87 | # 5 | |||||
| KC | 87.70 | # 4 | |||||
| Change Detection | DSIFN-CD | HANet | F1 | 62.67 | # 5 | ||
| IoU | 45.64 | # 4 | |||||
| Overall Accuracy | 85.76 | # 4 | |||||
| Precision | 56.52 | # 2 | |||||
| Recall | 70.33 | # 4 | |||||
| KC | 54.01 | # 2 | |||||
| Change Detection | GoogleGZ-CD | HANet | F1 | 75.28 | # 1 | ||
| Precision | 78.58 | # 4 | |||||
| Recall | 72.25 | # 4 | |||||
| Overal Accuracy | 88.34 | # 4 | |||||
| KC | 67.67 | # 4 | |||||
| IoU | 60.36 | # 4 | |||||
| Change Detection | LEVIR+ | HANet | F1 | 77.56 | # 4 | ||
| Prcision | 79.70 | # 3 | |||||
| Recall | 75.53 | # 4 | |||||
| OA | 98.22 | # 4 | |||||
| KC | 76.63 | # 4 | |||||
| IoU | 63.34 | # 4 | |||||
| Change Detection | LEVIR-CD | HANet | F1 | 90.28 | # 16 | ||
| IoU | 82.27 | # 11 | |||||
| Overall Accuracy | 99.02 | # 7 | |||||
| F1-score | 90.28 | # 4 | |||||
| Precision | 91.21 | # 5 | |||||
| Recall | 89.36 | # 6 | |||||
| Change Detection | S2Looking | HANet | F1-Score | 58.54 | # 7 | ||
| Precision | 61.38 | # 4 | |||||
| Recall | 55.94 | # 3 | |||||
| OA | 99.04 | # 4 | |||||
| KC | 58.05 | # 4 | |||||
| IoU | 41.38 | # 4 | |||||
| F1 | 58.54 | # 3 | |||||
| Change Detection | SYSU-CD | HANet | F1 | 77.41 | # 1 | ||
| Precision | 78.71 | # 3 | |||||
| Recall | 76.14 | # 2 | |||||
| OA | 89.52 | # 3 | |||||
| KC | 70.59 | # 4 | |||||
| IoU | 63.14 | # 4 | |||||
| Change Detection | WHU-CD | HANet | F1 | 88.16 | # 5 | ||
| Overall Accuracy | 99.16 | # 5 | |||||
| Precision | 88.30 | # 4 | |||||
| Recall | 88.01 | # 4 | |||||
| KC | 87.72 | # 4 | |||||
| IoU | 78.82 | # 4 |
