摘要
由于古城墙加固图像的成像光路复杂,各图像特征区域相互依赖交织,使得生成的古城墙结构信息准确性不足,三维测量结果的准确度下降。为了缓解这一问题,本文提出了最大池化与自注意力池化相结合的古城墙加固工程三维测量技术。首先,通过分析古城墙成像光路的光线中心,建立傅里叶变换窗口,经过三步移相重构古城墙图像,去除图像噪声;其次,结合最大池化方法与自注意力池化机制,对古城墙图像进行双重池化操作,计算每个图像在自注意力池化层中的权重,解析图像中各特征区域之间的依赖关系,从而生成古城墙图像的关键结构特征;最后,根据生成的图像特征对古城墙三维模型进行参数标定,实现古城墙加固的三维测量。依托工程实践可知,该项技术在实践应用中的置信区间值(CI)达到了0.75以上,表现出了较高的三维测量精度,满足了古城墙加固工程中的数据测量需求。
Due to the complex imaging optical path of the ancient city wall reinforcement images,the image feature regions are interdependent and intertwined,leading to insufficient accuracy in the generated structural information of the ancient city wall and a decline in the accuracy of the three-dimensional(3D)measurement results.To address this issue,this paper proposed a three-dimensional measurement technology for ancient city wall reinforcement that combines max pooling and self-attention pooling.First,by analyzing the center of the light path in the imaging of the ancient city wall,a Fourier transform window was established,and the ancient city wall image was reconstructed through three-phase shifting to remove image noise.Then,by combining the max pooling method with the self-attention pooling mechanism,a dual pooling operation was performed on the ancient city wall image.The weight of each image in the self-attention pooling layer was calculated to analyze the dependency relationships between the feature regions in the image,thereby generating the key structural features of the ancient city wall image.Finally,based on the generated image features,the parameters of the ancient city wall 3D model were calibrated to achieve 3D measurement for the ancient city wall reinforcement.According to engineering practice,the confidence interval(CI)of this technology in practical applications reaches above 0.75,demonstrating high accuracy in 3D measurement and meeting the data measurement requirements in the ancient city wall reinforcement project.
作者
李国勇
LI Guoyong(Guangdong Mingyuan Survey and Design Company Limited,Heyuan,Guangdong 517000,China)
出处
《北京测绘》
2025年第5期636-641,共6页
Beijing Surveying and Mapping
基金
广东省重点领域研发计划(2020B0101130009)。
关键词
自注意力池化机制
双重池化操作
古城墙加固
傅里叶变换窗口
三维测量
self-attention pooling mechanism
dual pooling operation
ancient city wall reinforcement
Fourier transform window
three-dimensional(3D)measurement
作者简介
李国勇(1993-),男,江西景德镇人,大学本科,工程师,研究方向为机器视觉与超声波传感技术和三维测量技术。E-mail:v99591817@163.com。
