摘要
利用传统视觉算法提取叶片边缘易受地平线或非目标风机的干扰对提取结果造成误差。针对这一问题,提出一种基于显著性及改进双向级联网络的检测算法,该算法可排除地平线及背景影响,获取完整边缘信息。其中,显著性检测网络模拟人的注意机制,提取视场重要信息,将目标风机与背景分离;改进双向级联网络在原有结构新增尺度选择模块,优化不同层之间特征共享,精确检测不同尺度的边缘,完整获取目标轮廓。利用某风电场无人机采集的实验数据,对识别性能评估得出该方法具有足够的精度,可在不同环境中准确识别叶片轮廓。
The traditional vision algorithm is used to extract the blade edge,which is easy to be interfered by horizon or non-target fan.To solve this problem,a detection algorithm based on saliency and improved Bi-Directional Cascade Network is proposed,which can eliminate the influence of horizon and background and obtain the complete edge information.The significance detection network simulates the human attention mechanism,extracts the important information of the field of view,and separates the target fan from the background.To improve the Bi-Directional Cascade Network,a new scale selection module is added to the original structure to optimize feature sharing between different layers,accurately detect the edges of different scales,and obtain the target contour completely.Based on the experimental data collected by UAV in a wind farm,the identification performance evaluation shows that the method has sufficient accuracy and can accurately identify the blade contour in different environments.
作者
张英
刘宾
Zhang Ying;Liu Bin(School of Information and Communication Engineering,North University of China,Taiyuan 030051,China)
出处
《国外电子测量技术》
北大核心
2023年第7期140-145,共6页
Foreign Electronic Measurement Technology
关键词
风机叶片边缘
显著性检测
双向级联
fan blade edge
significance detection
bidirectional cascade
