STF-EGFA: A Remote Sensing Spatiotemporal Fusion Network with Edge-Guided Feature Attention

Spatiotemporal fusion in remote sensing plays an important role Earth science applications by using information complementarity between different data to improve image performance. However, several problems still exist, such as edge contour blurring and uneven pixels the predicted real ground image, extraction of salient features convolutional neural networks (CNNs). We propose a spatiotemporal method with edge-guided feature attention based on sensing, called STF-EGFA. First, module is used maintain details, which effectively solves boundary problem. Second, make adaptive adjustments extracted features. Among them, spatial mechanism solve problem weight variation channels network. Additionally, pixel distribution addressed (PA) highlight transmit encoder (FA) at same time after union. Furthermore, weights edges, pixels, other are adaptively learned. Finally, three datasets, Ar Horqin Banner (AHB), Daxing Tianjin, verify method. Experiments proved that proposed outperformed typical comparison methods terms overall visual effect five objective evaluation indexes: spectral angle mapper (SAM), peak signal-to-noise ratio (PSNR), correlation coefficient (SCC), structural similarity (SSIM) root mean square error (RMSE). Thus, algorithm feasible for analysis.