小麦籽粒硬度的高光谱图像无损检测研究

发布时间：2018-04-13 22:07

本文选题：高光谱图像 + 麦粒硬度　；参考：《华北水利水电大学》2017年硕士论文

【摘要】：小麦作为我国的主粮,其品质的好坏直接影响到我们日常饮食生活的安全,且其籽粒硬度是评价谷物品质的一个重要参数。因此,如何能快速、有效、准确地检测麦粒硬度具有非常重要的研究意义。本文以温麦6号、中麦895号和西农979号3种不同硬度的品种为样本,研究了基于近红外高光谱图像的不同硬度麦粒可区分的理论机理,提取了能够表征麦粒硬度的有效区域,并对22类不同品种麦粒的近红外高光谱图像数据进行了预处理,建立了基于径向基核函数极限学习机回归分析技术的智能测定模型,实现了对小麦籽粒硬度的自动无损检测。(1)基于PCA的麦粒硬度的高光谱图像无损检测机理研究利用高光谱图像采集系统对22类品种的1540个麦粒进行图像采集,并依据国标法分别测定不同品种的麦粒实际硬度值。通过图像预处理提取出可有效表征麦粒硬度的感兴趣区域ROI,并以麦粒ROI内的平均光谱曲线作为该麦粒的近红外特征光谱。经预处理分析后将原始特征波长从871.6-1766.3nm范围内的256个减少到902.1-1699.6nm内的232个有效波长。利用PCA法对麦粒的高光谱图像进行分析,其中PC1、PC2和PC3的贡献率之和超过99.15%,故选取前3个主成分即可有效表征麦粒硬度的原始图像信息。依据不同硬度麦粒的像素点分布区域的密集程度可解释小麦籽粒样本间的化学差异,最终利用PC2和PC3组合的得分图研究麦粒硬度分类的可行性。建立了基于偏最小二乘判别分析的分类验证模型,正确识别率为100%,证实了基于PCA的麦粒硬度的高光谱图像无损检测机理研究的可行性。(2)基于人工蜂群优化算法的小麦籽粒硬度的波长选择针对麦粒近红外高光谱三维数据的特征波段多、混合度大及信息冗余多等特点,利用人工蜂群优化算法对特征波长进行优化选择。对于ABC算法中存在的早收敛、易陷入局部最优及接近全局最优解时搜索速度缓慢等缺点,提出了基于混沌搜索思想的人工蜂群优化算法。结果表明:从902.1-1699.6nm波段范围内的232个波长中优选出902.1-935.9nm、968.7-992.6nm、1042.7-1072.4nm等范围内的105个波长子集,波长数据量减少了54.7%。与ABC算法相比,CABC优化算法的运行时间缩短了45.3%,MSE降低了0.042%,SCC增加了0.55%。(3)基于RBF-ELM的小麦籽粒硬度智能预测模型为解决ELM算法中人工设置参数带来的系统不稳定性问题,利用GSM自动确定RBF-ELM算法中参数的输入,并基于RBF-ELM的回归分析技术建立小麦籽粒硬度的智能测定模型。结果表明:RBF-ELM预测模型与ELM模型相比,运行速度差别不大,但在精度和相关系数上分别提高了3.31%和7.36%;且与SVR模型相比,在训练和预测时间上均缩短了三个数量级,预测精度降低了0.53%,相关系数提高了0.96%。因此,利用基于RBF-ELM回归分析技术的智能测定模型有效实现了麦粒硬度的自动无损检测。
[Abstract]:As the main grain in China, the quality of wheat directly affects the safety of our daily diet, and its grain hardness is an important parameter to evaluate the grain quality.Therefore, it is very important to study how to detect wheat hardness quickly, effectively and accurately.In this paper, three varieties with different hardness, Wenmai 6, Zhongmai 895 and Xinong 979, were used as samples to study the theoretical mechanism of differentiating wheat grains with different hardness based on near infrared hyperspectral images, and to extract the effective regions which could characterize the hardness of wheat grains.The near infrared hyperspectral image data of 22 different varieties of wheat were preprocessed, and an intelligent measurement model based on radial basis function (RBF) learning machine regression analysis technique was established.The mechanism of hyperspectral image nondestructive testing of wheat grain hardness based on PCA was realized. The hyperspectral image acquisition system was used to collect the image of 1 540 wheat grains of 22 varieties.According to the national standard method, the actual hardness values of wheat grains of different varieties were determined.The region of interest (ROI), which can effectively characterize the hardness of wheat grains, was extracted by image preprocessing, and the average spectral curve in ROI was used as the near infrared characteristic spectrum of wheat grains.After pretreatment, the original characteristic wavelength was reduced from 256 in 871.6-1766.3nm range to 232 effective wavelengths in 902.1-1699.6nm.The hyperspectral images of wheat grains were analyzed by PCA method. The total contribution rate of PC1 + PC2 and PC3 was more than 99.150.Therefore, the original image information of wheat hardness could be effectively characterized by selecting the first three principal components.The chemical difference among wheat grain samples can be explained according to the density of pixel distribution area of wheat grains with different hardness. Finally, the feasibility of wheat hardness classification was studied by using the score chart of PC2 and PC3 combination.A classification verification model based on partial least squares discriminant analysis is established.The correct recognition rate is 100%, which confirms the feasibility of studying the mechanism of nondestructive detection of wheat grain hardness based on PCA.) the wavelength selection of wheat grain hardness based on artificial bee colony optimization algorithm is aimed at the near infrared hyperspectral triplet of wheat grain.There are many characteristic bands of dimensional data.Because of large mixing degree and much information redundancy, artificial bee colony optimization algorithm is used to optimize the selection of characteristic wavelengths.For the shortcomings of early convergence, easy to fall into local optimum and slow search speed when approaching the global optimal solution in ABC algorithm, an artificial bee colony optimization algorithm based on chaotic search idea is proposed.The results show that 105 subsets of the wavelength of 902.1-935.9nmhmhmnmnmnmnmnmhl9442.7-1072.4 nm are selected from the 232 wavelengths in the range of 902.1-1699.6nm band, and the amount of wavelength data is reduced by 54.7%.Compared with ABC algorithm, the running time of CABC-based optimization algorithm is shortened by 45.3MSE and 0.042%. SCC increases 0.55%.) the intelligent prediction model of wheat grain hardness based on RBF-ELM can solve the problem of system instability caused by manual setting parameters in ELM algorithm.The input of parameters in RBF-ELM algorithm was automatically determined by GSM, and an intelligent measurement model of wheat grain hardness was established based on RBF-ELM regression analysis technique.The results show that compared with the ELM model, the operation speed of the weight RBF-ELM model is not different from that of the ELM model, but the accuracy and correlation coefficient are increased by 3.31% and 7.36%, respectively, and the training and prediction time are shortened by three orders of magnitude compared with the SVR model.The prediction accuracy is reduced by 0.53, and the correlation coefficient is increased by 0.96.Therefore, an intelligent model based on RBF-ELM regression analysis was used to realize the automatic nondestructive testing of wheat grain hardness.
【学位授予单位】：华北水利水电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S512.1;TP391.41

【参考文献】