【作者】 吴迪；

【导师】 何勇；

【作者基本信息】 浙江大学 ， 生物系统工程， 2011， 博士

【摘要】 微藻在能源、食品工业、生物技术、医药工业、动物饲料、环境检测、污染治理等领域拥有广阔的开发应用前景。微藻具有生长繁殖快、光合作用效率高、倍增时间短、单位面积产量高等优点。随着能源、粮食、环境问题的日益突出,微藻生物质产业的发展已经得到了世界各国的广泛重视。目前微藻生物质产业中遇到的关键问题之一是如何提高微藻生产效率、增加产品经济价值并降低生产成本。为此本文提出“数字微藻”的概念。数字微藻的实现,即通过高科技手段,在微藻育种、养殖、收获和产品加工销售等环节快速、准确地获取微藻相关品质信息,用于品质监控,进行系统优化,最终实现微藻数字化科学生产与管理,从而提高微藻生产效率、保证质量、降低成本,增加经济收益,其中微藻品质信息的快速准确获取是实现微藻数字化科学生产与管理的基础和关键。数字微藻必将成为世界微藻生物质产业发展的研究重点和热点。而传统微藻信息获取方法耗时费力,不适合微藻生物质产业现代化生产的发展需要。本论文应用显微成像、光谱分析、高光谱成像、核磁共振等技术,结合多种图像处理技术和化学计量学算法,针对微藻育种、养殖、收获和产品加工销售等环节对相关品质信息的需要,研究了微藻形态、生命和品质信息的快速获取理论和方法,为微藻生物质产业高效生产和系统优化奠定了基础。本论文的主要研究内容和成果如下：(1)提出了微藻藻丝形态特征快速提取方法。通过对藻丝显微图像进行预处理,获得垂直朝向的单一藻丝二值化图像,以便藻丝形态特征参数提取；采用离散曲线演化技术(DCE)技术获取藻丝特征顶点；提出了一个新的算法用于去除骨架枝权,并获取了藻丝长度信息。该算法能够解决传统骨架提取算法在提取藻丝骨架时容易引入枝权的不足；实现了包括螺旋度、藻丝宽度、螺径、螺距、螺环数和紧密度等藻丝形态特征参数的快速获取。螺旋程度测量的平均误差为4.7%,藻丝宽度测量的的平均误差为6-2%,螺径测量的的平均误差为5.6%；开发了螺旋藻形态特征快速提取软件。该软件与人工测量方法比检测时间从5分钟减少到30秒,检测精度从93%提高到99%。(2)首次应用光谱和高光谱成像技术,建立了光谱和高光谱图像信息与微藻生命信息的定量关系模型,实现了微藻主要生命信息的快速获取。光谱检测中,透射光谱测量法和透反射光谱测量法的检测能力要优于反射光谱测量法和反应器外部光谱测量法；基于光谱技术的干物质重、油脂单位体积含量和油脂单位质量含量最优模型的rpre2分别达到0.9836,0.9777和0.9487；光谱特征变量模型平均变量个数为9.41个。和全波段模型相比有99.62%的变量被去除,而模型rpre2平均值仅下降4.19%;12个光谱模型中有8个模型的的UVE-SPA特征变量选择结果要优于直接进行SPA计算。说明LIVE可以有效提高SPA对光谱特征变量提取的效率；基于高光谱成像技术的干物质重、叶绿素单位体积含量、叶绿素单位质量含量、叶绿素a单位体积含量、叶绿素a单位质量含量、叶绿素b单位体积含量、叶绿素b单位质量含量最优模型的rpre2分别达到0.9891、0.9882、0.9242、0.9895、0.9444、0.9780和0.9282；高光谱图像特征变量模型平均变量个数减少到了7.43个(99.7%的变量被去除),而rpre2平均值则达到0.9550(全波段模型为0.9573)；7个高光谱光谱图像模型中有5个的特征变量最优提取算法为UVE-SPA。说明UVE可以有效提高SPA对高光谱光谱图像信息特征变量提取的效率；基于高光谱图像,获得了微藻生命信息检测指标的藻液分布图。结果表明高光谱成像技术在微藻生命信息获取能力上要明显优于RGB图像。(3)研究构建了藻油ω-3多不饱和脂肪酸含量快速获取方法和系统。核磁共振技术获得了最优的藻油(o-3多不饱和脂肪酸(DHA和EPA)含量检测结果。模型rval2,分别为0.9625和0.9674；可见-短波近红外光谱、长波近红外光谱和和中红外光谱最优检测模型对藻油DHA含量检测的屹7分别为0.9190、0.9232和0.8748,对EPA含量检测的吃,最大值分别为o.9213、0.8757和0.8857,但均未能达到核磁共振技术的检测精度；基于可见光激光源的拉曼光谱技术不能准确测量藻油DHA和EPA含量。其模型RMSECV分别高达21.2707和1.8529,是核磁共振谱模型RMSECV值的1.67和2.68倍；特征变量选择能有效提高藻油DHA和EPA含量检测精度,模型RMSECV平均值能够降低18.70%和29.03%；10个藻油品质指标检测模型中有9个的最佳特征变量选择算法为UVE结合SPA,说明UVE可以有效提高SPA在藻油品质指标检测过程中特征变量提取的效率。(4)研究建立了藻粉品质信息快速获取方法与模型。研究了藻粉类别的可见-近红外光谱快速检测方法。基于7个特征波长的SPA-LS-SVM模型分类正确率达到100%。研究了藻粉蛋白质含量的可见-近红外光谱快速检测方法。UVE-SPA-MLR为最佳检测模型,其中rpre2、RMSEP和RPD分别为0.9750、0.2344和6.2206,检测精度能够满足实际要求；研究了藻粉辐照剂量的光谱快速检测方法。UVE-SPA-BP-ANN模型为最优的藻粉辐照剂量检测模型,其中rpre2、RMSEP和RPD分别为0.9850、0.6414和8.1366,检测精度能够满足实际要求；研究了藻粉掺假信息的光谱快速检测方法。DVA分析结果表明短波近红外光谱比全波段光谱更适合被用于藻粉掺假信息的快速获取,而LS-SVM算法要优于PLS和PLS2算法。对于单一掺假物、两组掺假物和多组掺假物情况下面粉和绿豆粉LS-SVM短波近红外光谱检测模型的rpre2分别为0.9966、0.9430和0.9740,以及0.9903、0.9474和0.9705。更多还原

【Abstract】 Microalgae have broad application prospects in many fields such as fuel, energy, food industry, biotechnology, pharmaceutical industry, animal feed, environmental monitoring, and pollution control. Microalgae have excellent qualities of fast growth and reproduction, high photosynthetic efficiency, short doubling time, high production per acre, low greenhouse gas emission and little or no competition with food production. Because of the growing threats of energy, food and environmental issues, microalgal biorescoure industry has gained more attention worldwide. At present, one of the key questions is how to improve the growing efficiency, increase the value of algal products and reduce the production cost. This thesis proposed a new concept of DIGITAL MICROALGAE, which means using high-tech methods to rapidly and non-invasively measure the microalgal growth information and qualities of algal products in the processes of microalgal breeding, cultivation, harvesting, processing and sales. The obtained information can be used for the optimal decision-making control and system operation, and ultimately optimizes the production and management of microalgae, and improves production efficiency, ensures the quality of products, reduces costs and increases economic benefits. The foundation and key in the process of Digital Microalgae is the rapid and non-invasive determination of microalgal growth information and qualities of algal products. Digital Microalgae will become the research priorities and hot spots of the world. At present, however, conventional off-line analyses are rather time-consuming and inefficient which can not meet the needs of modern production of microalgal biorescoure industry.In this work, according to the requirements of relevant quality information of microalgal in the processes of breeding, cultivation, harvesting, processing and sales, microscopy imaging technique, spectral analysis technique, hyperspectral imaging technique, nuclear magnetic resonance (NMR) technique combined with image process algorithms and chemometrics were used for rapid and non-invasive determination of comprehensive quality information of microalgal such as microscopic morphological features, growth information, qualities of algal oils and algal powders to further provide information support for the high efficient production and system optimization of microalgal biorescoure industry. The main research contents and results are shown as follow:(1) Microscopy image processing algorithms were proposed to rapid measure morphological features of Spirulina microalga filaments. After the image preprocessing, binary images of each filament at vertical position were obtained. The filament rotation was applied to make the filament characteristic parameters extraction easier. The vertices of filaments were determined based on discrete contour evolution (DCE) algorithm. Because traditional algorithm of skeleton extraction has the disadvantage of bringing in branches, a new algorithm was proposed to reduce the branches and obtain the Spirulina length based on the main skeleton. Morphological features such as length, diameter of helix, degree of spiralization, width of filament, pitch, helix number, and helix pitch were obtained based on the vertices of filaments. The errors between manually obtained and automatic calculated values were 4.7%for degree of spiralization,5.6%for diameter of helix and 6.2%for width of filament, respectively. Furthermore, a Spirulina Morphological Feature Extraction System Software was developed. The feature extraction time is about 30 seconds and the measurement accuracy was 99%by using the software, while manual measurement usually takes five minutes with the accuracy of 93%.(2) Spectroscopy and hyperspectral imaging techniques were for the first time applied to establish the quantitative relationship models between the spectra and hyperspectral image information and algal growing information respectively and realized the rapid determination of algal growing information. In the spectral analysis, transmittance model and transflectance model were better than reflectance model and reflectance measured outside the bioreactor model. The best coefficients of determination of prediction (rpre2e)values of the models were 0.9836,0.9777 and 0.9487 for the spectral analysis of the dry weight, lipid content per unit volume and lipid content per unit weight. There were average 9.41 variables for the spectral efficient variable models. Compared to the whole spectral models,99.62%of the variables were eliminated, while the average rpre2 values of the spectral efficient variable models only decreased 4.19%. In all the twelve spectral efficient variable models, for eight models, uninformative variable elimination combined with successive projections algorithm (UVE-SPA) performed better than using SPA directly in the variable selection processes. It shows that UVE could effectively improve the variable selection accuracy of SPA. The rpre2 values of the hyperspectral image models for the measurement of dry weight, chlorophyll content per unit volume, chlorophyll content per unit weight, chlorophyll a content per unit volume, chlorophyll a content per unit weight, chlorophyll b content per unit volume and chlorophyll b content per unit weight were 0.9891,0.9882,0.9242,0.9895,0.9444, and 0.9780 respectively. There were avaerage 7.41 variables for the hyperspectral image efficient variable models. Compared to the whole spectral models,99.7% variables were eliminated, while the average rpre2 value of the spectral efficient variable models was 0.9550 (0.9573 for the whole spectral model). In all the seven hyperspectral image efficient variable models, for five models, UVE-SPA was better than using SPA directly in the variable selection processes. It shows that UVE could improve the hyperspectral image variable selection accuracy compared to using SPA directly. Moreover, the quality distribution maps of microalgal slurry were obtained based on the algal hyperspectral images. The results show that hyperspectral imaging technique is better than RGB images to measure the growing information of algae.(3) Rapid determination method and system of the content of w-3 PUFAs in algal oil were proposed. NMR obtained the best DHA and EPA prediction models. The best coefficients of determination of validation (rpre2) values of two models were 0.9625 and 0.9674 respectively. The best rpre2, values of visible and short-wave near infrared spectral model, long-wave near infrared spectral model, and mid-infrared spectral model were 0.9790,0.9232 and 0.8748 for DHA analysis,0.9213,0.8757 and 0.8857 for EPA analysis respectively. However their results were not as good as those of NMR models. The Raman spectroscopy with 514 nm light laser didn’t perform well for DHA and EPA prediction. The RMSECV of their models were 21.2707 and 1.8529, which were 1.67 and 2.68 times of those of NMR models. The effective variable selection can improve the DHA and EPA prediction accuracy. Models’average RMSECV values decreased 18.70%and 29.03%respectively. In all the ten efficient variable models, for nine models, UVE-SPA was better than using SPA directly in the variable selection processes. It shows that UVE could improve the variable selection accuracy compared to using SPA directly in the algal oil analysis.(4) Rapid determination method and models of the quality of algal powders were established. Based on visible and near infrared spectroscopy, SPA-least square support vector machine (SPA-LS-SVM) with seven effective variables reached 100%correct answer rate for the algal powder classification. Visible and near infrared spectroscopy was used to predict the protein content in algal powders. UVE-SPA-multiple linear regression (UVE-SPA-LS-SVM) obtained the best result with the rpre2, root mean square error of prediction (RMSEP) and residual predictive deviation (RPD) of 0.9750,0.2344, and 6.2206, respectively. The detection accuracy can meet the practical requirements. Spectroscopy technique was used to determine the irradiation dose of algal powders. UVE-SPA-Back-Propagation Artificial Neural Network (UVE-SPA-BP-ANN) obtained the best result with the rpre2, root mean square error of prediction (RMSEP) and residual predictive deviation (RPD) of 0.9850,0.6414, and 8.1366, respectively. The detection accuracy can meet the practical requirements. Visible and near infrared spectroscopy was used to predict the adulterant contents in algal powder. The design value analysis (DVA) indicated that for quantification of adulterants in algal powder, short-wave near infrared spectroscopy outweighs full spectra, and LS-SVM models outweigh PLS and PLS2 models. The rpre2 values of LS-SVM models based on NIR spectra were 0.9966,0.9430, and 0.9740 for flour detection and 0.9903,0.9474, and 0.9705 for mung-bean detection, while single adulterant, two adulterants and multiple adulterants were considered.更多还原