OFA-NET: Optical Flow Aligning Network for Time Series Change Detection

With the rapid advancement of remote sensing and Earth observation technologies, a plethora of Time Series remote sensing Images (TSIs) from platforms like Landsat and Sentinel-2 have become accessible, providing valuable data for Time Series remote sensing images Change Detection (TSCD). In TSCD, TSIs captured at the same geographic location but at different times often exhibit misalignment issues due to variations in radiation incidence angles, satellite orbit deviations, and other factors. To tackle these challenges, an Optical Flow Aligning Network (OFA-NET) for TSCD is proposed in this paper. In detail, a simple and lightweight Encoder-Decoder is introduced to extract multiscale spatial features from TSIs and restore these features to the size of the original input of Encoder-Decoder. Moreover, Encoder-Decoder employs the Siamese structure to effectively preserve the original features of each input image as much as possible. Subsequently, the outputs of Encoder-Decoder are utilized for optical flow prediction, which is further employed to align TSIs and extract differences between them for TSCD. Experiments conducted on the UTRNet datasets illustrate that the proposed OFA-NET achieves superior accuracy and produces more distinct change detection results compared to other methods.