【作者】 李细荣；

【导师】 胡永彪；

【作者基本信息】 长安大学 ， 机械设计及理论， 2013， 博士

【摘要】 土的压实度是工程建设基础施工最重要的质量指标，检测工作量大，并要求准确、快速、方便、实时，使得土的压实度无损检测成为压实度检测研究的热点问题。本文利用光电技术和计算机技术，对基于激光图像的土的压实度无损检测检测方法进行了探讨研究。鉴于土的类别和结构成分的多样性和复杂性，本文选取粘土为研究对象，假定土的密实度是均值。主要工作如下：首先，论文对土的压实度检测技术和激光图像技术相关内容的研究现状及其存在的主要问题进行了归纳和总结，指出了土的压实度检测方法存在的不足，以及激光图像检测土压实度的存在问题；从土的含水率测量方法、最佳含水率和最大干密度等方面论述了土压实度的评价方法以及压实理论，系统分析了压实度的影响因素。对比分析了土组织和生物组织结构的特点，借鉴激光在生物组织中传输规律及激光成像理论来研究激光在土组织中传播过程及激光成像。分析了土组织的光学特性以及光子在组织中的辐射传输方程，并且利用漫射近似理论对传输方程求解；根据激光在土组织中的传输规律及成像理论，通过土组织激光图像找出与其压实度相关的特征。对基于激光图像的土压实度的检测系统的基本原理、系统组成和检测的关键技术进行了研究分析。其次，对能否利用漫射近似理论对传输方程求解得出土的光学参数问题进行了研究。提出用蒙特卡罗模拟方法验证漫射近似理论，并分析了蒙特卡罗方法及模拟过程，然后在假定土的光学参数的前提下，用C语言编制程序模拟了光在土组织中的传播过程以及得出了漫反射率，并通过漫射近似理论计算出漫反射率，并将两者结果进行了比较分析，两者结果的最大绝对误差为0.0497和最大相对误差为10.44%。结果表明：由于蒙特卡罗方法对任意的反照率、测量位置和边界条件都是成立的，所以可认为蒙特卡罗模拟结果是足够精确的。证明了用漫射近似理论对辐射传输方程求解得出土的光学特性参数是可行的。然后，对土组织激光图像的试验和结果进行了研究。依据击实试验制备了土组织试样，通过激光图像检测系统采集了激光图像；通过现有图像处理方法找到了图像中心位置，针对现有图像处理方法的处理步骤多和速度慢的问题，提出了Matlab Gui界面交互处理方式，减少了处理步骤以及提高了速度。并分析了不同含水率的土组织表面漫反射光的分布规律；通过漫射理论方程和最小二乘法计算出土组织的光学参数。然后分析了漫反射率的变化率、光学参数和图像灰度均值变化率与压实度的相关性，漫反射率变化率随着压实度的增大而有增大的趋势，灰度变化率随着压实度的上升而有下降的趋势；吸收系数随着压实度的增大有减小的趋势，散射系数随压实度增大有增大的趋势。同时分析了激光图像的现有纹理特征提取算法，现有算法提取的均匀性、能量和第3阶矩特征随着压实度的增大而增大的趋势，而其相关度、对比度、平均亮度、平均对比度、平滑度、一致性和熵特征随着压实度的增大而减小的趋势；并提出了新的纹理特征算法，新算法提取的相关和逆差矩特征随着压实度的增大而增大的趋势；而其小梯度优势、大梯度优势、灰度分布的不均匀性、灰度平均、惯性、梯度平均、灰度均方差和梯度均方差特征随着压实度的增大而减小的趋势。为了比较两种纹理特征提取算法，下一章节分别将两种算法提取的特征建立预测模型，比较其预测精度。最后，对BP神经网络预测压实度和试验验证问题进行了研究。针对与土压实度的相关特征较多的问题，提出了神经网络方法对压实度进行预测。我们选取所有与压实度相关的特征作为模型的输入变量，分别将漫反射率变化率、灰度均值变化率、吸收系数、散射系数、对比度、平均亮度、一致性、均匀性、能量、相关度、第三阶矩值、平均对比度、平滑度和熵共14个特征作为第一组原始变量；而漫反射率变化率、灰度均值变化率、吸收系数、散射系数、相关、逆差矩、小梯度优势、大梯度优势、灰度分布的不均匀性、灰度平均、梯度平均、灰度均方差、梯度均方差和惯性共14个特征作为第二组原始变量。并通过主成分分析，将原始数据中减少为5个主成分因子，然后分别利用这两组特征数据建立预测模型，并进行预测，将预测结果与环刀法结果比较，第一组特征的预测值的平均绝对误差为0.0937和平均相对误差为10.08%，第二组特征的预测值的平均绝对误差为0.0714和平均相对误差为7.71%；第二组特征的预测值比第一组特征的预测精度高，因此，本文使用第二组特征所建立的预测模型，且其预测精度表明，用BP神经网络模型预测土的压实度是可行的。最后在实验条件不变的情况下，采集另外不同的土样数据，对所建立的BP神经网络预测模型进行了验证，验证后的平均绝对误差为0.0862和平均相对误差为8.76%，结果表明：在实验条件不变、研究对象为粘土、假定密实度是均值的情况下，同时受到试验的操作误差和激光图像噪声等因素的影响下，可认为利用激光图像技术检测土的压实度是可行的。更多还原

【Abstract】 The degree of compaction of the soil is the most important indicator of quality ofconstruction foundation construction.However,as the testing work is heavy and needs to beaccurate, fast, convenient and real-time, therefore the research of non-destructive testing ofthe degree of compaction of the soil becomes a hot issue.In this paper, the non-destructivetesting detection method of the degree of compaction of the soil which is based on the laserimage will be researched by using the photovoltaic technology and computer technology. Themain tasks are as follows:First, in this paper, the current research situation and the main problems of testingtechnology of the degree of compaction of soil and laser imaging technology are summedup, and the shortcomings of the method of testing soil compaction are also pointed out. Theevaluation method of soil compaction and compaction theory are discussed from the aspectsof measurements of soil moisture content, optimum moisture content and maximum drydensity and the main factors of soil compaction are systematically analysed.And similaritiesand differences of the organizational structure of indigenous organizations and biological arecomparatively analysed. laser propagation in the soil organization and laser imaging arestudied by drawing on the experience of the theory of laser transmission rule and laserimaging in biological tissue. The paper also analyses the optical properties of the soilorganization and photonic radiative transfer equation in the organization and uses the epitaxialboundary conditions diffusion approximation theory for solving the transport equation.According to the law of the transmission of the laser in the soil organization and its imagingtheory and through soil tissue laser image, the optical parameters of the soil organizationsrelated to their degree of compaction and the change rate of laser image gray value can befound out.The basic principles of the laser image-based detection systems of soil compactiondegree, system components and the key technologies of testing are studied and analyzed.Secondly, I did some research about whether the diffusion approximation theory can beapplied to the transfer equation solving and obtain the optical parameters of soil. The paperanalysed Monte Carlo method and simulation, then used the Monte Carlo method to simulatethe propagation of light in the soil organization,and finally analysed and compared thesimulation results with the diffusion approximation theory results.The result of thecomparison shows: the maximum average absolute error is0.0497and the maximum average relative error is10.44%between two. The results show that using the radiation propagationtheory and diffusion theory to obtain the optical properties of soil is feasible.Then, the paper studied the test and results of the laser image on the soil tissue.According to the soil tissue sample obtained through the compaction test, and in accordancewith the law of the transmission of the laser in the soil organizations and imaging theory,the laser image is collected by using the laser image detection system through the soilorganizations;The location of the center of the image can be found out through existingimage processing method. aimed at the problems about the processing steps and slow-speedof existing image processing method, the paper put forward the Matlab GUI interfaceinteraction approach,which reduces the processing steps as well as improve the speed. And thedistribution of diffuse light on the surface of the soil tissue of different moisture content isalso analysed. The experimental results show that the scattering and distribution of light in thesoil tissues is mainly decided by the optical characteristics of the soil organization; theabsorption and scattering coefficients of the soil tissue can be calculated by using thediffusion theory equation and the least-squares method. And then the change rate of thediffuse reflectance of the soil tissue, absorption coefficient, scattering coefficient, The changerate of the laser image gray value and correlation of soil compaction degree are analysed,Diffuse reflectance change rate increases with the increase of the degree of compaction,Grayrate of change has a downward trend with the rise of the degree of compaction.Absorptioncoefficientu atends to decrease as the degree of compaction increases.Scattering coefficientu shas an increasing trend as the degree of compaction increases. At the same time,the existingtexture features algorithm of the laser image is analysed and new image texture featuresalgorithm is put forward.The result is that the uniformity of image texture, energy, the3rdorder moment value, related and inverse difference moment have a tendency to increase withthe increase of the degree of compaction.However,small gradient strengths, gradientadvantage, the uneven distribution of gray, gray average, average gradient gray variance,gradient variance, inertia correlation, contrast, average brightness, average contrast,smoothness, consistent and entropy value decrease with the increase of the degree ofcompaction trend.Finally, the paper studies how to use the BP neural network to predict the degree ofcompaction and test verification.Because of the so many characteristics of the degree of soil compaction,therefore, the neural network prediction and evaluation of the degree ofcompaction is raised. We selected all characteristics related to the degree of compaction asinput variables of the neural network model. The diffuse reflectance rate of change, the graymean rate of change, the absorption coefficient, scattering coefficient, contrast, an averageluminance of consistency, uniformity, energy, correlation, the third order moment value, theaverage contrast, smoothness and entropy are treated as the first set of original variables.Thediffuse reflectance rate of change the gray mean rate of change, the absorption coefficient,scattering coefficient, related deficit moment, the advantages of small gradients, gradientstrengths, the uneven distribution of gray, gray average gradient average grayvariance,gradient variance and inertia are treated as a second set of the original variables.And throughthe principal component analysis (PCA), the original data was reduced to five principalcomponents factor.The contribution rate of the main component factors of characteristic datain the first group is97.3%,The contribution rate of the main component factors ofcharacteristic data in the second group is98.6%, and then respectively use these two sets offeature data to establish BP neural network prediction model to predict the degree ofcompaction of soil samples and compare the predicted results with those of Cutting Ringmethod.The average absolute error of the predicted value of characteristic data in the firstgroup was0.0937and the average relative error is10.08%.The average absolute error of thepredicted value of characteristic data in the second group is0.0714and the average relativeerror is7.71%. the predicted value of characteristic data in the second group is moreaccurate than those of the first set.So,this paper uses the characterization data in the secondset to establish prediction model.And the prediction accuracy shows that using the BP neuralnetwork model to predict soil compaction is feasible.The BP neural network prediction modelwas also verified by a different soil data.The verifIied average absolute error is0.0862and theaverage relative error of8.76%,which proves that using laser imaging to detect soilcompaction is feasible.更多还原