基于无人机图像地面建筑物目标检测算法研究

Research on the Algorithm for Detecting Objects on the Ground Based on Images by UAV

利用无人机图像基于深度学习对地面建筑物目标进行检测,特别是针对偏僻地区房屋目标的检测,相对人工方式提高了检测效率。应用无人机航拍技术获取视频数据,建立专用的建筑物数据集,提出一种YOLOv5算法的改进模型,融合注意力机制并改进损失函数,添加针对小目标的检测层。实验结果表明,建筑物目标识别的准确率为94.2%,平均精确度均值为96.4%,检测速度为每秒88.9帧。

Using images captured by unmanned aerial vehicle to detect building targets on the ground,especially houses targets in remote areas,has a higher detection efficiency than manual methods.In this paper,UAV aerial photography technology is applied to acquire video data,and special building data set is established.An improved version of deep learning-based YOLOv5 algorithm is proposed.Attention mechanism is integrated,loss function is improved,and small target detection layer is added.The experimental results show that the accuracy of building target recognition is 94.2%,the average accuracy is 96.4%,and the detection speed is 88.9 frames per second.