摘要
为提高大坝裂缝检测的效率和精度,解决传统大坝裂缝检测技术人工劳动强度大、存在人身安全隐患的问题,本文提出一种基于图像视觉信息的人工智能裂缝检测技术,设计了基于注意力机制的语义分割网络,在U形网络中引入了多头注意力机制网络结构,将移动网络V3编码器提取的特征图在多尺度上经多头注意力机制模块特征强化与解码器结果进行融合输出,精确识别裂缝对象,采用数学形态学腐蚀膨胀操作计算裂缝的几何特征。实验表明,得到的裂缝图像轮廓清晰、连续无锯齿、走向与实际相符,具有较好的检测精度。本文方法在检测大坝裂缝时省时省力、不受地形环境制约、部署方便,在准确识别大坝裂缝以及对大坝裂缝病险进行综合评判时具有一定价值。
In order to improve the efficiency and accuracy of dam crack detection while addressing the issues of high manual labor intensity and personal safety risks associated with traditional methods,this paper proposes an artificial intelligence crack detection technique based on image visual information.It introduces an attention mechanism-based semantic segmentation network with a multi-head attention mechanism network structure incorporated into the U-shaped network.This approach enhances the features extracted by the MobileNetV3 encoder at multiple scales using multi-head attention mechanism modules and fuses them with the decoder results to accurately identify crack objects.Mathematical morphological operations,such as erosion and dilation,are applied to calculate the geometric characteristics of the cracks.Experimental results demonstrate that the obtained crack image contours are clear,continuous without jagged edges,and consistent with reality,showcasing good detection accuracy.This method is efficient and convenient for detecting dam cracks,is not constrained by the terrain environment,and is easy to deploy.It holds certain value in accurately identifying dam cracks and providing comprehensive assessments of potentially hazardous dam cracks.
作者
王泽焘
WANG Zetao(GEPIC Hexi Hydropower Development Co.,Ltd.,Zhangye 734000,China)
出处
《水利建设与管理》
2024年第2期39-44,共6页
Water Conservancy Construction and Management
关键词
大坝裂缝检测
注意力机制
特征提取
形态学
dam crack detection
attention mechanism
feature extraction
morphology
