水果组织光学特性参数反演模型及其应用研究

发布时间：2018-11-07 08:26

【摘要】：水果组织光学特性参数是反映水果组织自身化学成分、物理结构和生理、病理状态的重要参数。水果组织光学特性参数的测量,对于研究组织内部结构成像规律和光子的传输特点,分析组织化学特性和物理结构,建立组织内部品质(状态)检测和评价模型等具有重要意义。本文基于MC仿真模拟了光在单层水果组织中的传输现象,并实现了光学特性参数的逆向求解;其次利用高光谱散射成像系统采集了组织模拟液的高光谱散射图像,结合建立的非线性反演回归模型,实现了组织模拟液光学特性参数的反演求解;并在此研究基础上,研究了苹果组织对光谱的吸收和散射情况,建立了光谱特征与苹果硬度和可溶性固体含量之间的预测模型。论文的主要工作如下:1.针对漫射模型与MC模拟在近光源处存在较大误差,提出了一种基于迭代反演的输运平均自由程估计及光源-检测器最小距离确定方法。该方法利用迭代估计思想,自适应地评估出输运平均自由程的值,并改变光源-检测器最小距离,从而获得较为合理的用于光学特性参数反演的数据区间。结果表明:与传统经验估计方法相比,迭代反演方法能够减少近光源处误差的引入,有效地提高水果组织光学特性参数的反演准确度。在无噪声的条件下,吸收系数μ_a反演的平均相对误差为7.17%;有效散射系数μ_s反演的平均相对误差为5.73%。在加入一定信噪比噪声的情况下,迭代反演方法仍然能获得较高的光学特性参数反演准确度。2.由于光学近似模型存在各种限制,研究中采用机器学习方法建立光学特性参数μ_a和μ_s的预测模型。利用基于稳态空间分辨技术的高光谱散射成像系统获取组织模拟液530-900nm波段范围内的散射图像,结合傅里叶分解和最小二乘支持向量机算法建立光学特性参数的非线性反演回归模型。结果表明:基于实验数据的傅里叶分解和最小二乘支持向量机的建模方法能获得更好的预测结果,μ_a和μ_s反演的平均相对误差分别为11.03%和7.16%。3.研究了苹果的硬度和可溶性固体含量(SSC)预测模型。在线高光谱散射成像系统用于采集2009 年和 2010 年的'Golden Delicious,(GD), 'Jonagold,(JG)和'Delicious'(RD)苹果样本500-1000nm波段范围内的散射图像。利用光学特性参数方法、矩方法和傅里叶分解方法分析高光谱散射图像并提取光谱特征,结合偏最小二乘和最小二乘支持向量机,建立苹果硬度和SSC的预测模型。结果表明:融合后的光谱特征(光学特性参数μ_a和μ_s、零阶矩和一阶矩、傅里叶系数)相比于单一光谱特征,能提供更多有关于散射曲线的信息,从而提高了苹果硬度和SSC的预测精度。
[Abstract]:The optical properties of fruit tissue are important parameters reflecting the chemical composition, physical structure, physiological and pathological state of fruit tissue. The measurement of the optical characteristic parameters of fruit tissue, for studying the imaging law of the inner structure of the tissue and the characteristics of photon transmission, analyzing the histochemical characteristics and physical structure, It is of great significance to establish internal quality (state) detection and evaluation model. In this paper, the transmission of light in single layer fruit tissue is simulated based on MC, and the inverse solution of optical characteristic parameters is realized. Secondly, the hyperspectral scattering images of tissue simulated fluid are collected by hyperspectral scattering imaging system, and the inversion solution of the optical characteristic parameters of tissue simulation fluid is realized by combining with the nonlinear inversion regression model. On the basis of this study, the absorption and scattering of apple tissue were studied, and the prediction model between the spectral characteristics and the hardness and soluble solid content of apple was established. The main work of this paper is as follows: 1. In view of the large errors between the diffuse model and the MC simulation near the light source, an iterative inversion based method for estimating the mean free path of transport and determining the minimum distance between the light source and the detector is proposed. The method adaptively evaluates the mean free path of transport by using iterative estimation idea and changes the minimum distance between light source and detector to obtain a more reasonable data interval for the inversion of optical characteristic parameters. The results show that compared with the traditional empirical estimation method, the iterative inversion method can reduce the near-light source error and improve the retrieval accuracy of the optical characteristic parameters of fruit tissue. Under the condition of no noise, the average relative error of absorption coefficient 渭 _ a inversion is 7.17 and the average relative error of effective scattering coefficient 渭 _ s inversion is 5.73. In the case of certain SNR noise, the iterative inversion method can still obtain higher inversion accuracy of optical characteristic parameters. 2. Due to the various limitations of the optical approximation model, the prediction models of the optical characteristic parameters 渭 _ a and 渭 _ s are established by using the machine learning method. The hyperspectral scattering imaging system based on steady-state spatial resolution technique is used to obtain the scattering images in the 530-900nm band range of tissue simulation fluid. The nonlinear inverse regression model of optical parameters is established by combining Fourier decomposition and least squares support vector machine (LS-SVM) algorithm. The results show that the method of Fourier decomposition and least squares support vector machine based on experimental data can obtain better prediction results. The average relative errors of 渭 _ a and 渭 _ s inversion are 11.03% and 7.16.3 respectively. The (SSC) prediction model of apple hardness and soluble solid content was studied. The online hyperspectral scattering imaging system is used to collect the scattering images of 'Golden Delicious, (GD),' Jonagold, (JG) and 'Delicious' (RD) apple samples in 500-1000nm band range from 2009 to 2010. The hyperspectral scattering images were analyzed and extracted by optical characteristic parameter method, moment method and Fourier decomposition method. The prediction model of apple hardness and SSC was established by combining partial least squares and least squares support vector machine. The results show that the fused spectral features (optical parameters 渭 _ a and 渭 _ s, zero-order moments and first-order moments, Fourier coefficients) can provide more information about the scattering curves than the single spectral features. Thus, the prediction accuracy of apple hardness and SSC is improved.
【学位授予单位】：江南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S66;TP391.41

【参考文献】