【摘要】 遥感图像存在着目标特征不明显、背景信息复杂、检测效率低下等问题。针对这些问题,本文提出了一种改进的SA-YOLOv5s遥感图像检测算法,将注意力模块ShuffleAttention加入YOLOv5骨干特征提取和颈特征融合网络的卷积块中,并在公开的RSOD遥感图像数据集上进行了实验。实验结果表明,改进后的算法平均精度为96.2%,较原有的YOLOv5s算法提升了1.1%。更多还原

【Abstract】 Remote sensing images have problems such as unclear target features, complex background information, and low detection efficiency. This article proposes an improved SA-YOLOv5s remote sensing image detection algorithm to address these issues. The Shuffle Atten-tion module was added to the convolutional blocks of the YOLOv5 backbone feature extraction and neck feature fusion network, and experiments were conducted on the publicly available RSOD remote sensing image dataset. The experimental results show that the average accuracy of the improved algorithm is 96.2%, which is 1.1% higher than the original YOLOv5s algorithm.更多还原