基于雷达点云与视觉图像融合的输电线路探鸟驱鸟技术

Bird Detecting and Bird Repelling Technology for Transmission Lines Based on the Fusion of Radar Point Cloud and Visual Image

为了克服驱鸟器易被鸟类适应的缺点以及解决基于视觉图像的目标检测算法受天气条件制约的问题,根据毫米波雷达的特性,提出了一种基于雷达点云与视觉图像融合的探鸟驱鸟方法。首先运用均值拟合实现雷达-相机坐标融合,并利用视觉和场景增强技术创建了囊括多种气象环境的鸟类点云-图像融合数据集;然后提出一种将雷达点云注意力机制与深度学习识别网络YOLO相结合的鸟类识别模型,实现决策层融合;最后,结合三帧差分算法构建基于3F-GIo U的驱鸟器智能启停策略,判断是否有鸟类停留在目标区域,使其适合于轮廓小、速度快的目标行为识别。实验结果表明,所提的鸟类识别方法能够满足实际应用场景中鸟类识别的鲁棒性和准确性,不同天气条件下平均识别准确度为91.21%;所提的3F-GIoU策略能够有效识别鸟类存在危害线路、杆塔的活动。

In order to overcome the shortcoming that the bird repellent is easy to be adapted by birds and to solve the problem that the target detection algorithm based on visual image is limited by weather conditions,a bird hunting and re-pelling method based on the fusion of radar point cloud and visual image is proposed according to the characteristics of millimeter-wave radar.Firstly,the radar-camera coordinate fusion is achieved by means of the mean fitting.Then,a bird point-cloud-image fusion dataset that covers a variety of meteorological environments is created by using vision and sce-ne enhancement techniques.Secondly,a bird recognition model that combines the radar point-cloud attention mechanism and the deep learning recognition network YOLO is proposed to realize the fusion of decision-making layers.Finally,the intelligent start-and-stop strategy based on 3F-GIoU is constructed by combining the three-frame difference algorithm to determine whether there are birds staying in the target area,which is suitable for the target behavior recognition with small outline and fast speed.The experimental results show that the bird identification method proposed in this study can meet the robustness and accuracy of bird identification in practical application scenarios,and the average identification accuracy under different weather conditions is 91.21%.Furthermore,the 3F-GIoU strategy proposed in this study can ef-fectively identify the activities of birds that endanger lines and towers.