基于改进Faster R-CNN的无人机航拍图像目标检测

Object Detection in UAV Aerial Images Based on Improved Faster R-CNN

无人机航拍图像中目标检测问题要求检测模型具有旋转不变性。针对这一问题,提出改进的 Faster R-CNN 算法。首先在区域建议网络中采用 K-means 聚类方法生成适应数据集的预设锚点框,其次在 Fast R-CNN网络中引入新的特征提取层,并在模型多任务损失函数中增加旋转约束条件,为后续检测学习旋转不敏感特征。在人工采集的数据集上进行了对比实验,结果表明:在检测速度无明显降低的情况下,改进方法的检测精度提升了1. 6%mAP,算法检测性能较优,更能满足实际应用需求。

To solve the problem of object detection in UAV aerial images,which requires an detector with rotation-invariance,this paper proposes an improved Faster R-CNN algorithm. Firstly,in region proposal network,K-Means clustering is used to generate pre-set anchors that enhance adaptability for different datasets. Then,in Fast R-CNN,it introduces a new feature layer and adds rotation constraints to the model’s multi-task loss function for learning rotation-insensitive feature.Experiments on collected dataset show that the improved method increases the accuracy by 1. 6% mAP without significant decline at speed,which demonstrates the improved method is better and more suitable for application.