
福州大学电气工程与自动化学院,福州,350108
Published:2025
移动端阅览
JIANG Hao, HE Hao, LIU Xinyu, et al. 基于多模态融合的输电线路鸟巢精细化检测方法[J]. 2025, (11): 5435-5447.
JIANG Hao, HE Hao, LIU Xinyu, et al. 基于多模态融合的输电线路鸟巢精细化检测方法[J]. 2025, (11): 5435-5447. DOI: 10.13336/j.1003-6520.hve.20241704.
输电线路杆塔上的鸟巢检测与评估,对预防输电线路故障、保障电力系统安全稳定运行具有重要意义。输电线路鸟巢的稳定性与其结构和形状特征密切相关,结构不对称或形状异常的鸟巢更容易发生巢材脱落,从而引发输电线路故障。目前,鸟巢故障风险评估主要基于通用目标检测和定位方法,尚未考虑鸟巢结构差异和形状特征的影响,导致评估结果不准。为此,提出一种基于多模态融合的输电线路鸟巢精细化检测方法,具体包括图像特征提取、语义信息提取和多模态融合3个模块。图像特征提取模块采用K-Net图像分割框架,精确分割鸟巢轮廓并裁剪图像,减少无关信息干扰,提取图像模态特征。语义信息提取模块着重提取与鸟巢下垂状态密切相关的形状和结构语义信息,并建立语义信息关联模型进行动态修正,以生成有效的文本模态特征。多模态融合模块采用低秩多模态融合策略LMF,利用文本模态对鸟巢结构和形状的语义理解弥补图像模态无法捕捉的细节特征,以生成更具表征能力的融合特征,实现鸟巢精细化检测。实验证明,该方法在目标检测与定位基础上,进一步分辨了鸟巢的下垂状态,检测精度达92.59%,为鸟巢故障风险评估和预防工作提供了可靠依据。
Detection and assessment of bird nests on transmission line towers are crucial for preventing faults in transmission lines and ensuring the safe and stable operation of power systems. The stability of the bird nests on transmission lines is closely related to its structural and shape characteristics
and bird nests with asymmetric structure or abnormal shape are more prone to material shedding
thus causing transmission line faults. Current risk assessments of bird nest faults primarily focus on object detection and localization
without fully accounting for the impact of the structural differences and shape characteristics of bird nests
resulting in inaccurate assessment results. To this end
a fine-grained detection method for bird nests on transmission lines based on multimodal fusion is proposed
comprising following three modules: image feature extraction
semantic information extraction
and multimodal fusion. The image feature extraction module utilizes the K-Net image segmentation framework to accurately segment the contours of bird nests and crop the images
reducing irrelevant information interference and extracting image modality features. The semantic information extraction module focuses on extracting shape and structural semantic information closely related to the sagging trends of bird nests
and a semantic correlation model for dynamic correction is established to generate effective text modal features. The multimodal fusion module adopts the Low-rank Multimodal Fusion (LMF) strategy
in which the semantic understanding of bird nest structure and shape from the text modality are utilized to compensate for the fine details that the image modality cannot capture
in order to generate more representative fusion features for fine-grained detection of bird nests. Experimental results demonstrate that this method can be adopted to further distinguish the sagging state of the bird nests on the basis of object detection and localization
achieving a detection accuracy of 92.59%
which provides a reliable basis for bird nest failure risk assessment and prevention work.
0
Views
0
下载量
0
CSCD
Publicity Resources
Related Articles
Related Author
Related Institution
京公网安备11010802024621