【摘要】 针对Faster-RCNN算法在遥感图像当中对小目标的漏检、检测精度不高等问题作出改进.用特征提取能力更强的ResNet50网络替换VGG16;同时为了加强对遥感小目标信息的提取，引入特征金字塔，添加多尺度扩张卷积模块来增强特征金字塔的上下文特征，扩充小目标信息，使用通道注意力机制来减少特征融合过程中带来的信息混淆，提高模型对遥感小目标的检测效果.实验表明，所改进的方法在HRRSD遥感数据集达到86.7%的检测精度，较改进前提升了5.2%,同时检测效果也优于当前的一些主流检测模型，证明了改进后模型的有效性.更多还原

【Abstract】 The Faster-RCNN algorithm has been improved to address the problems of missing small objects and low detection accuracy in remote sensing images.In order to enhance the extraction of remote sensing small object information, we introduced a feature pyramid network, and added a multi-scale expansion convolution module to enhance the contextual features of the feature pyramid network, expand the small object information, and use the channel attention mechanism to reduce the information confusion caused by the feature fusion process and improve the detection of remote sensing small objects.The experiments show that the improved method achieves 86.7% detection accuracy on the HRRSD remote sensing dataset, which was 5.2% higher than that before the improvement, and also outperforms some current mainstream detection models, proving the effectiveness of the improved model.更多还原