Cross-Dataset Generalization of Compact Building Change Detectors: From LEVIR-CD to SpaceNet 7
Main Article Content
Keywords
building change detection, cross-dataset generalization, LEVIR-CD, SpaceNet 7, source-only transfer
Abstract
Within-dataset scores do not establish performance after sensor, geography, and label definitions change. We compare Change Vector Analysis (CVA) with three compact learned controls under a common LEVIR-CD training regime, then freeze checkpoints, source RGB moments, and validation-selected thresholds before external evaluation. The external set contains 13 SpaceNet 7 areas of interest (AOIs), each with fixed 10- and 22-month pairs, for 26 pairs in total. On LEVIR-CD, the learned controls achieved F1 scores of 0.6541-0.7425, compared with 0.1398 for CVA. The ranking reversed on SpaceNet 7: CVA reached 0.1740 macro-F1, whereas the learned controls reached 0.0334-0.0650. Their pooled AUPRC and target-oracle F1 also remained low, so threshold recalibration alone could not recover the source advantage. An eightfold scale-degradation proxy reduced learned LEVIR-CD F1 by 0.6229-0.7197. A label-free target-statistics radiometric control improved all three learned controls, but it did not belong to the strict source-only result. These findings define a protocol-specific generalization boundary for the evaluated compact controls and support external validation before building-change maps are used as auxiliary evidence for urban-stock monitoring.
References
- [1] Daudt, R. C., Saux, B. L., and Boulch, A., Fully convolutional siamese networks for change detection, 2018 IEEE international conference on image processing (ICIP), 4063–4067, 2018, https://doi.org/10.1109/ICIP.2018.8451652.
- [2] Chen, H., and Shi, Z., “A spatial-temporal attention-based method and a new dataset for remote sensing image change detection,” Remote Sensing, 12, (10), 1662, 2020, https://doi.org/10.3390/rs12101662.
- [3] Chen, H., Qi, Z., and Shi, Z., “Remote sensing image change detection with transformers,” IEEE Transactions on Geoscience and Remote Sensing, 60, 1–14, 2022, https://doi.org/10.1109/TGRS.2021.3095166.
- [4] Etten, A. V., Hogan, D., Martinez-Manso, J., Shermeyer, J., Weir, N., and Lewis, R., The multi-temporal urban development SpaceNet dataset, Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 6398–6407, 2021, https://doi.org/10.1109/CVPR46437.2021.00633.
- [5] Shen, L., Lu, Y., Chen, H., Wei, H., Xie, D., Yue, J., Chen, R., Lv, S., and Jiang, B., S2Looking: A satellite side-looking dataset for building change detection, 2021, https://arxiv.org/abs/2107.09244v3.
- [6] Townshend, J. R. G., Justice, C. O., Gurney, C., and McManus, J., “The impact of misregistration on change detection,” IEEE Transactions on Geoscience and Remote Sensing, 30, (5), 1054–1060, 1992, https://doi.org/10.1109/36.175340.
- [7] Fang, S., Li, K., Shao, J., and Li, Z., “SNUNet-CD: A densely connected siamese network for change detection of VHR images,” IEEE Geoscience and Remote Sensing Letters, 19, 1–5, 2022, https://doi.org/10.1109/LGRS.2021.3056416.
- [8] Chen, J., Yuan, Z., Peng, J., Chen, L., Huang, H., Zhu, J., Liu, Y., and Li, H., “DASNet: Dual attentive fully convolutional siamese networks for change detection in high-resolution satellite images,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2020, https://doi.org/10.1109/JSTARS.2020.3037893.
- [9] Wang, Z., Peng, C., Zhang, Y., Wang, N., and Luo, L., “Fully convolutional siamese networks based change detection for optical aerial images with focal contrastive loss,” Neurocomputing, 2021, https://doi.org/10.1016/j.neucom.2021.06.059.
- [10] Bandara, W. G. C., and Patel, V. M., A transformer-based siamese network for change detection, 2022 IEEE international geoscience and remote sensing symposium, 207–210, 2022, https://doi.org/10.1109/IGARSS46834.2022.9883686.
- [11] Liu, J., Xuan, W., Gan, Y., Zhan, Y., Liu, J., and Du, B., “An end-to-end supervised domain adaptation framework for cross-domain change detection,” Pattern Recognition, 132, 108960, 2022, https://doi.org/10.1016/j.patcog.2022.108960.
- [12] Corley, I., Robinson, C., and Ortiz, A., A change detection reality check, ICLR workshop on machine learning for remote sensing, 2024, https://arxiv.org/abs/2402.06994.
- [13] Radke, R. J., Andra, S., Al-Kofahi, O., and Roysam, B., “Image change detection algorithms: A systematic survey,” IEEE Transactions on Image Processing, vol. 14, no. 3, pp. 294–307, 2005, doi: 10.1109/TIP.2004.838698.
- [14] Tewkesbury, A. P., Comber, A. J., Tate, N. J., Lamb, A., and Fisher, P. F., “A critical synthesis of remotely sensed optical image change detection techniques,” Remote Sensing of Environment, vol. 160, pp. 1–14, 2015, doi: 10.1016/j.rse.2015.01.006.
Article Sidebar
This work is licensed under a Creative Commons Attribution 4.0 International License.