PSSANet: Phase-Sensitive and Scale-Adaptive Anchor-Free framework for Geospatial object detection

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Remote sensing object detection is critical for environmental monitoring, maritime surveillance, and urban management. However, detecting arbitrarily oriented objects in aerial images remains challenging due to large scale variations, dense distributions, and complex backgrounds. To address these challenges, this paper proposes an anchor-free oriented object detection framework named PSSANet, which integrates three key components: a lightweight LRFormer backbone for enhanced global contextual modeling, a Scale-and-Channel-Aware Feature Pyramid Network to improve multi-scale feature representation through adaptive scale-aware aggregation, and an Adaptive Phase-Shifting Coder Head that introduces structural modulation and phase-based angle encoding to achieve more stable and accurate orientation prediction. Experiments on the DOTA-v1.0 dataset show that PSSANet achieves a mAP of 75.21 %, representing a 3.91 % improvement over the baseline Rotated FCOS. The method shows improvements on challenging categories, including helicopter (+21.6 % AP), harbor (+8.8 % AP), bridge (+5.7 % AP), and storage tank (+4.7 % AP), indicating stronger capability in detecting small and complex objects. Furthermore, the model achieves 70.9 % mAP on the RSAR dataset and 89.90 % AP50 on the HRSC2016 dataset, demonstrating robustness and cross-dataset generalization ability. These results confirm that the proposed framework improves detection accuracy of rotated objects in complex aerial scenarios while maintaining adaptability across different remote sensing datasets.