A Spatial–Temporal Difference Aggregation Network for Gaofen-2 Multitemporal Image in Cropland Change Area
Food security is an important guarantee of peace and development in the world. The accurate monitoring of cropland utilizing remote sensing data provides a strong technical support for the protection of cropland resources. Nonetheless, in contrast to the building change detection, the growth charact...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10813395/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Food security is an important guarantee of peace and development in the world. The accurate monitoring of cropland utilizing remote sensing data provides a strong technical support for the protection of cropland resources. Nonetheless, in contrast to the building change detection, the growth characteristics of crops in cropland areas exhibit significant variations in accordance with different seasonal climates and light intensities. Furthermore, the serious imbalance between the cropland change area and the nonchange area makes it difficult to focus on the real change area in the cropland under these interferences. To this end, we propose a spatial–temporal difference aggregation network (STDAN) for cropland change detection (CCD), which can focus on the real change area between different temporal images. Specifically, we use a cross-temporal difference feature enhancement module to enhance the difference features while establishing the correlation between different temporal features, which can suppress task-independent interference. Subsequently, the cross-level difference feature aggregation (CDFA) realizes the aggregation between different levels of difference features in an incremental manner to further refine the change area. Finally, the utilization of multireceptive fusion enables the integration of different scale characteristics obtained by CDFA, thereby yielding the accurate CCD outcomes. The experimental results indicate that the proposed STDAN achieves the highest <italic>F</italic>1, IOU, OA, and Kappa scores at 79.63%, 66.16%, 97.05%, and 78.04%, respectively, on the Gaofen-2 cropland data. In addition, we conduct generalization experiments on the remaining three mainstream datasets, demonstrating that our method is equally applicable to other change detection scenarios. |
|---|---|
| ISSN: | 1939-1404 2151-1535 |