Dual-Attention-Driven Multiscale Fusion Object Searching Network for Remote Sensing Imagery

Object search is a challenging yet important task. Many efforts have been made to address this issue and achieve great progress in natural image, searching all specified types of objects from remote sensing image barely studied. In work, we are interested images. Compared person scenes, task two factors: One that usually contains large number objects, which poses challenge characterize the object features. Another images dense, easily yields erroneous localization. To these issues, propose new end-toend deep learning framework for First, multi-scale feature aggregation (MSFA) module, strengthens representation lowlevel features by fusing multi-layer The fused with richer details significantly improve accuracy search. Second, dual-attention enhancement (DAOE) module enhance channel spatial dimensions. enhanced localization dense objects. Finally, built datasets based on images, complex changes space time. experiments comparisons demonstrate state-of-the-art performance our method datasets.