基于遥感图像的多尺度特征增强方法与研究
摘要
算法。该算法在Backbone部分采用SPPF-MSA多尺度特征增强模块,代替了原始YOLOv11的SPPF,可以提升多尺
度特征表达能力。在Neck中提出分层差异化MSA-GCNet特征融合策略,增强不同尺度目标的特征表示。最后,在
head层引入基于CSL角度编码的OBB旋转检测头,缓解了角度回归过程中梯度发散问题,提升旋转目标角度预测精
度。实验结果表明,与原始YOLOv11n算法相比,在平均精度和准确率上分别提升2.7%和4.2%。
关键词
全文:
PDF参考
[1]Ren S, He K, Girshick R, et al. Faster R-CNN:
Towards real-time object detection with region proposal
networks[C]//Proceedings of the IEEE International Conference
on Computer Vision. Santiago: IEEE, 2015: 91-99.
[2]Xia G S, Bai X, Ding J, et al. DOTA: A large-scale
dataset for object detection in aerial images[C]//Proceedings
of the IEEE Conference on Computer Vision and Pattern
Recognition. Salt Lake City: IEEE, 2018: 3974-3983.
[3]Ding J, Xue N, Long Y, et al. Learning ROI transformer
for oriented object detection in aerial images[C]//Proceedings
of the IEEE Conference on Computer Vision and Pattern
Recognition. Long Beach: IEEE, 2019: 2849-2858.
[4]Yang X, Yan J, Yang X, et al. R3Det: Refined singlestage detector with feature refinement for rotating object
detection[C]//Proceedings of the AAAI Conference on Artificial
Intelligence. New York: AAAI, 2021, 35(4): 3163-3171.
[5]杨雪,严俊驰. Arbitrary-oriented object detection
with circular smooth label [C]// 欧洲计算机视觉会议组委
会.2020欧洲计算机视觉大会论文集。柏林:欧洲计算机
视觉协会,2020:637-644
Refbacks
- 当前没有refback。
