摘要
针对传统目标检测算法对航拍影像中的车辆等小目标检测效果差的问题,提出一种基于改进Faster R-CNN的无人机地物车辆检测算法。该算法在原始Faster R-CNN网络的基础上融合了FPN作为基础网络模型——FFRCNN,采用ResNet-50代替原始VGG-16作为主要骨干网络进行多特征融合,使用Focal Loss损失函数改善正负样本不均衡的情况。在改进网络的基础上,使用空洞卷积将多尺度空间的特征信息进行融合,提高网络的感受野,更好地收集图像的上下文信息。实验结果表明,改进后的检测算法平均精确度达到93.8%,相较于原FFRCNN网络,平均精确度提升了19.2%,具有更好的鲁棒性。
Aiming at the problem that the traditional target detection algorithm has poor detection effect on small vehicle targets in aerial imagesa vehicle detection algorithm based on improved Faster R-CNN is proposed.Based on the original Faster R-CNN networkthis method combines FPN as the basic network model—FFRCNNResNet-50 is applied instead of original VGG-16 as the main backbone network for multi-feature fusionand Focal Loss function is used to correct the imbalance between positive and negative samples.On the basis of improving the networkatrous convolution is used to fuse the feature information of multi-scale spacewhich can improve the receptive field of the network and better collect the context information of the image.The experimental results show that the average accuracy of the improved detection algorithm reaches 93.8%which is 19.2%higher than that of the original FFRCNN network and has better robustness.
作者
宋建辉
王思宇
刘砚菊
于洋
池云
SONG Jianhui;WANG Siyu;LIU Yanju;YU Yang;CHI Yun(School of Automation and Electrical Engineering,Shenyang Ligong University,Shenyang 110000,China;Liaoning Provincial Party School of CPC,Shenyang 110000,China)
出处
《电光与控制》
CSCD
北大核心
2022年第7期69-73,80,共6页
Electronics Optics & Control
基金
国家重点研发计划基金资助项目(2017YFC0821001)。