Remote Sensing Object Detection Based on Fusion of Spatial and Channel Attention

Remote sensing object detection faces unique challenges due to objects' varied scales and orientations. To address these challenges, we propose the Spatial Channel Attention Fusion Module (SCAF-Module), designed to enhance detection accuracy by integrating multi-scale convolutions, adaptive rotated convolutions, and parallel spatial channel attention mechanisms. The experiments, conducted using the DOTA-vl.0 and HRSC2016 datasets, demonstrate the efficacy of the SCAF-Module. We achieved mean Average Precision (mAP) scores of 80.94% and 98.23% on these datasets, respectively. Comparative experiments reveal that the SCAF -Module surpasses several advanced models, including the baseline Oriented R-CNN. Additionally, ablation studies highlight the significance of the spatial and channel attention mechanisms and the impact of rotated convolutions on detection performance. The SCAF-Module presents a robust and adaptable framework for remote sensing object detection, offering significant improvements over existing methods. This work paves the way for further optimization and application of the module in other challenging remote sensing tasks.