摘要
现有输电线图像目标分割方法多基于传统图像处理或单一深度学习模型,在复杂背景和光照波动下易出现断裂、误分等问题,导致分割精度受限。因此,文章提出高空输电线舞动视觉图像目标鲁棒分割方法。文章通过动态噪声过滤、多尺度对比度提升和改进Sobel算子抑制背景干扰并强化输电线特征,构建融合残差结构和空间注意力模块的改进U-Net网络,基于强化特征引导输电线初始分割,利用自适应形态学运算和三次贝塞尔曲线拟合优化初始分割结果,通过拓扑关系分析和动态边界校准实现鲁棒分割。实验结果表明,该方法在整体测试集及复杂干扰场景下的交并比(Intersection over Union,IoU)分别达到0.896和0.851,显著优于对比方法,该方法具有更高的分割精度。
Existing image segmentation methods for power lines predominantly rely on traditional image processing techniques or single deep learning models,which often suffer from issues like segmentation fractures and misclassifications under complex backgrounds and lighting variations,resulting in limited accuracy.To address these challenges,this study proposes a robust segmentation method for moving high-voltage power line images.The approach employs dynamic noise filtering,multi-scale contrast enhancement,and an improved Sobel operator to suppress background interference while enhancing line features.A refined U-Net architecture integrating residual structures and spatial attention modules is constructed to guide initial segmentation through feature refinement.Adaptive morphological operations combined with cubic Bezier curve fitting are then applied to optimize initial segmentation results.Thus,robust segmentation is achieved through topological relationship analysis and dynamic boundary calibration.Experimental results demonstrate that the proposed method achieves a high Intersection-Union Ratio(IoU)of 0.896 on the full test dataset and 0.851 in complex interference scenarios,significantly outperforming comparative methods with superior segmentation accuracy.
作者
贵童
赵子昂
贾丽
GUI Tong;ZHAO Zi’ang;JIA Li(State Grid Henan Electric Power Company Zhoukou Power Supply Company,Zhoukou 466000,China)
出处
《无线互联科技》
2025年第22期105-108,共4页
Wireless Internet Science and Technology
关键词
高空输电线
视觉图像
目标分割
特征引导
形态学
high-voltage power lines
visual images
target segmentation
feature guidance
morphological analysis
