级联卷积网络在无人机视觉定位的应用

Application of Cascade Convolution Network in UAV Vision Location

针对当前无人机视觉定位精度对特征提取要求高,易受环境光影影响等问题,结合无人机视觉定位的特点和目标检测网络的优势,论文创新性提出了级联卷积网络。通过修改Faster R-CNN网络结构,设计级联R-CNN、语义级联和级联激活函数,将级联卷积网络取代传统图像处理模块引入无人机视觉定位算法,提高了目标特别是小目标的检测精度,检测结果可以达到87.9%。在实地无人机试飞实验中,该网络的定位误差在0.3m以内,满足无人机视觉定位的要求。

In view of the problems of the high demand for feature extraction and the environment on the UAV vision location,combined with the features of UAV vision position and the advantages of the object detection network,the cascade convolution network is innovatively proposed in this paper.By modifying the structure of Faster R-CNN network,cascade R-CNN,semantic cascade and cascade activation functions are designed.The cascade convolution network is replaced by the traditional image processing module to introduce the UAV vision localization algorithm,which improves the detection precision of the target,especially the small target,and the detection results can reach 87.9%.Through the actual UAV flight test,the positioning error of the network is within 0.3 m,which meets the requirements of UAV vision location.