摘要
原子力显微镜(AFM)因具备亚纳米级测量分辨力,成为形貌表征的关键工具。然而,针尖卷积效应会导致特征边缘模糊,制约AFM表征精确度。文章提出一种融合注意力增强机制的编解码深度学习算法。该算法在传统算法基础上引入通道注意力机制,优先处理针尖-表面作用效果强烈的特征通道,提高了针尖卷积伪影图像校正的准确度。为验证该算法有效性,以金纳米和钯纳米颗粒为研究对象,构建了符合其特征的、包含随机圆形与随机方形结构的纳米颗粒仿真图像数据集。仿真结果表明:相较于传统编码解码算法,改进算法在金纳米颗粒图像的训练任务中,结构相似性指数(SSIM)和边缘锐度系数(ESC)分别提高至0.9843和22.1081;在钯纳米颗粒图像的训练任务中,其峰值信噪比(PSNR)提高至30.77 dB,ESC提高至28.0715。改进算法有效提高了AFM测量精确度。
Atomic force microscopy(AFM)has become a key tool for topographical characterization due to its sub-nanometer measurement resolution.However,the tip-convolution effect leads to blurred feature edges,which compromises measurement accuracy.This study proposes a deep learning encoding-decoding algorithm based on an attention-augmented mechanism.Building upon conventional approaches,this algorithm incorporates a channel attention mechanism that prioritizes processing feature channels exhibiting strong tip-surface interactions,thereby enhancing the accuracy of tip-convolution artifact correction.To validate its effectiveness,a simulated dataset of nanoparticle images was constructed using gold and palladium nanoparticles as research subjects,featuring structures consistent with their characteristics-random circular and random square patterns.Simulation results demonstrate that compared to traditional encoding-decoding algorithms,the proposed algorithm achieves superior performance:for gold nanoparticle image training tasks,the Structural Similarity Index(SSIM)and Edge Sharpness Coefficient(ESC)improved to 0.9843 and 22.1081,respectively,and for training tasks involving palladium nanoparticle images,the Peak Signal-to-Noise Ratio(PSNR)increased to 30.77 dB,and the ESC rose to 28.0715.This algorithm effectively enhances the precision of AFM measurements.
作者
王艳艳
黄远婷
步明臻
黄俊龙
WANG Yanyan;HUANG Yuanting;BU Mingzhen;HUANG Junlong(School of Automation and Electrical Engineering,Tianjin University of Technology and Education,Tianjin 300222,China;Tianjin Key Laboratory of Information Sensing&Intelligent Control,Tianjin 300222,China)
出处
《天津职业技术师范大学学报》
2025年第4期1-7,共7页
Journal of Tianjin University of Technology and Education
基金
天津市自然科学基金面上项目(23JCYBJC01070)。
关键词
原子力显微镜
针尖卷积伪影
注意力增强机制
测量精确度
atomic force microscopy
tip convolutional artifacts
attention-augmented mechanism
measurement accuracy
