摘要
针对遥感图像的复杂特征、多尺度问题,对YOLOv5s算法进行了改进。首先引入Global Context Block模块并和主干网络中的C3模块相结合形成了融合全局上下文注意力的G3GC模块,增强主干网络的特征提取能力。然后,引入双向特征金字塔网络BiFPN对颈部网络做进一步优化升级,提高了模型对多尺度变化目标的检测能力。最后,使用动态非单调聚焦机制的WIOU损失函数代替CIOU损失函数,提高了预测框的精度。将改进后的算法在NWPU VHR-10 Dataset数据集上进行实验,实验结果表明,与YOLOv5s算法相比,改进算法的查准率、召回率、平均精度分别提高了4%、2.6%、2.1%。
In view of thecomplex features and multi-scale issues of remote sensing images,the YOLOv5s algorithm has been improved.Firstly,the Global Context Block module is introduced and combined with the C3 module in the backbone network to form the G3GC module fusedwith globalcontext attention,which enhances the feature extraction ability of the backbone network.Then,BiFPN is introduced to further optimize and upgrade the neck network,which improves the detection ability of the model for multi-scale changing targets.Finally,the WIOU loss function of the dynamic non-monotonic focusing mechanism is used instead of the CIOU loss function to improve the accuracy of the prediction box.Theimproved algorithm is tested on NWPU VHR-10 Dataset,and the experimental results show that the precision,recall and average precision of the improved algorithm are improved by 4%,2.6% and 2.1%,respectively,compared with YOLOv5s algorithm.
作者
梁攀攀
孔红山
刘彬
LIANG Pan-pan;KONG Hong-shan;LIU Bin(School of Cryptographic Engineering,Information Engineering University,Zhengzhou 450000,China)
出处
《计算机仿真》
2025年第11期331-335,364,共6页
Computer Simulation
关键词
深度学习
目标检测
遥感图像
全局上下文注意力
Deep learning
Object detection
Remote sensing image
Global context attention
