基于1D-ResNet的沥青混合料光谱分类识别方法  

Method of asphalt mixture spectral classification and recognition based on 1D⁃ResNet

作　　者：王晋军 周兴林[1] WANG Jinjun;ZHOU Xinglin(School of Mechanical and Automation,Wuhan University of Science and Technology,Wuhan 430081,China)

机构地区：[1]武汉科技大学机械自动化学院,湖北武汉430081

出　　处：《现代电子技术》2025年第8期139-144,共6页Modern Electronics Technique

基　　金：国家自然科学基金项目(51778509);国家科学自然基金项目(51827812)。

摘　　要：使用近红外光谱技术对沥青混合料的老化程度进行快速有效评估,对于沥青道路养护具有重要意义。为了实现不同老化程度沥青混合料的快速准确分类,提出一种基于一维残差卷积神经网络(1D-ResNet)的沥青混合料光谱分类方法。该方法是在卷积神经网络链式结构的基础上引入残差模块来构建1D-ResNet分类模型。首先对近红外光谱数据间隔平均,并进行二阶导数(2nd D)及标准正态变量变换(SNV)预处理;然后将归一化的平均光谱、2nd D光谱及SNV光谱进行光谱序列融合;最后将融合光谱数据作为模型的输入,实现对不同老化程度沥青混合料的分类。实验结果表明:对光谱数据进行间隔平均后,1D-ResNet模型分类准确率为88.38%,采用光谱序列融合后分类准确率达98.86%,能够实现对沥青混合料的准确分类识别。Using near-infrared spectroscopy technology to quickly and effectively evaluate the aging degree of asphalt mixture is of great significance for asphalt road maintenance.In order to realize the fast and accurate classification of asphalt mixtures with different aging degrees,a method of asphalt mixtures spectral classification based on one-dimensional residual convolutional neural network(1D-ResNet)is proposed.In this method,the residual module is introduced on the basis of the chain structure of convolutional neural network to construct the 1D-ResNet classification model.The interval of near-infrared spectral data is averaged and preprocessed with second derivative(2nd D)and standard normal variable(SNV).The spectral sequence fusion is performed on normalized average spectra,2nd D spectra,and SNV spectra.The fused spectral data is used as input for the model to realize the classification of asphalt mixtures with different degrees of aging.The experimental results show that the classification accuracy of 1D-ResNet model is 88.38%after interval averaging of spectral data,and the classification accuracy is 98.86%after spectral sequence fusion,which can accurately realize the classification and recognition of asphalt mixture.

关 键 词：沥青混合料 光谱分类 一维残差卷积神经网络 光谱预处理 序列融合 间隔平均法 

分 类 号：TN247-34[电子电信—物理电子学]