SIFNet: A self-attention interaction fusion network for multisource satellite imagery template matching

• DOI: 10.1016/j.jag.2023.103247
• Publication Date: 2023-03-03T11:16:16Z
• Authors: Ming Liu; Gaoxiang Zhou; Lingfei Ma; Liangzhi Li; Qiong Mei
• Keywords: Robustness; Template matching

Multisource satellite images provide abundant and complementary earth observations, while nonlinear radiometric geometric distortions (such as scale rotation variations) between these multimodal pose remarkable challenges for further remote sensing applications, such change detection. We therefore proposed a template matching algorithm based on self-attention interactive fusion network, named SIFNet, to align multisource images. First, feature pyramid network was first conducted extract multiscale features each pixel, with the reference inputs. Then, extracted were fused by layers in Transformer information interaction. Third, similarity semantic loss functions developed convert imagery registration into regression task, allowing SIFNet aligning patch more efficiently point-to-point correspondence, instead of globally searching extremums previous strategies did. performed experiments four datasets (i.e. Google, GF-2, Landsat-8 optical-SAR images) various scenes evaluate performance robustness SIFNet. The results demonstrate comparable accuracy other algorithms robust variations data.