【作者】 王大成；

【导师】 黄敬峰； 王纪华；

【作者基本信息】 浙江大学 ， 农业遥感与信息技术， 2011， 博士

【摘要】 生态因子是影响小麦品质形成的重要因素,本论文通过综合利用生态因子和遥感分区进行小麦籽粒蛋白质含量遥感监测,以期提高监测精度。以北京地区冬小麦为研究对象,开展以下研究：(1)基于神经网络的冬小麦蛋白质含量关键生态因子的筛选分析,针对影响冬小麦主要品质--蛋白质的百分含量,利用神经网络算法,确定了各个因子对品质影响力的大小并将其定量化表达；(2)利用所筛选的决定小麦品质关键生态因子,基于各个因子分别与其权重叠加的结果与各个因子不考虑权重叠加的结果两种分区模型进行区划研究,选择最优的区划模型实现了北京地区优质冬小麦品质等级分类的区划；(3)针对不同的分区构建综合生态因子和遥感信息的小麦品质监测模型,并与单一遥感监测模型、生态因子模型对比,选择最优的小麦品质监测模型,以期实现北京地区小麦品质的遥感监测。本研究对于增强小麦品质遥感监测机理、提高精度具有重要意义。研究取得的主要成果表现在以下几个方面：1.基于神经网络的冬小麦蛋白质含量关键生态因子初步分析利用北京地区具有代表性的小麦种植点的气象数据和土壤养分数据,通过神经网络方法来评估温度、降雨、光照和土壤养分含量等因子对小麦籽粒蛋白质含量影响的相对重要程度。研究表明,影响北京地区冬小麦籽粒蛋白质含量的主要因素依次有：蜡熟初期(6月6日至6月10日)的光照时间、气温大于32℃的天数、土壤碱解氮含量、整个灌浆期即籽粒形成关键期(5月上旬至6月上旬)的平均气温、灌浆期后期(5月26日至5月30日)的平均气温、灌浆期中后期(5月下旬至6月上旬)≥0℃的积温、乳熟期(6月1日至6月5日)的平均气温、灌浆期中后期(5月下旬至6月上旬)的温差、灌浆期中后期(5月下旬至6月上旬)的降雨量和土壤有机质含量；针对关键因子利用神经网络模型制作了响应曲线以反映蛋白质含量随生态因子的变化趋势。2.基于遥感与地理信息系统的北京地区冬小麦品质分区的研究在ARCGIS支持下,利用影响冬小麦蛋白质含量的各个关键生态因子进行空间插值,将点状关键因子数据空间化,建立多因子空间数据库,根据神经网络对每一因子计算出的RATIO设定一个权重值多层叠加,在ENVI环境下,用气象因子和土壤因子分层分析、整合各种因素,比较等权重和差异权重两种分区模型的分区结果,最终作出北京地区冬小麦品质分区的各种区域的不同分类；并对分类结果进行了分析。3.基于分区的小麦品质形成关键期蛋白质含量遥感监测以北京地区的主推冬小麦品种--中优206籽粒蛋白质为研究对象,利用遥感数据提取北京地区冬小麦不同生育时期多种植被指数(VIs)和中优206籽粒蛋白质进行相关性研究,结果表明：5月11日的NDVIgreen值与籽粒蛋白质相关性最好且达极显著水平,因此该时期为建立冬小麦遥感品质监测模型的最佳时相。利用生态环境数据和光谱数据分别构建了冬小麦光谱品质模型、生态环境品质模型以及光谱与生态环境综合品质模型；通过对冬小麦光谱品质模型、生态环境品质模型以及光谱与生态环境综合品质模型预测效果进行F检验,表明各模型均达到极显著水平；但是与其它两种模型相比,光谱与生态环境综合品质模型的决定系数(R2)有明显的提高,并且相对均方根误差(RRMSE)和相对误差(RE)降低,且降幅度较大。说明光谱与生态环境综合品质模型比单纯的生态环境品质模型和光谱品质模型有较好的预测效果。再进一步的五类分区建立模型的研究中,细分的模型各个精度都比整体模型精度有不同程度的提高,其中最优质区域预测精度为91.6%,二类区89.3%,三类区85.6%,四类区83.6%,品质最差区域为92.2%；对比发现,分区域建立模型在预测精度上比整体模型预测分别提高了12%、10.5%、7.3%、11.7%和14.3%。因此,利用遥感和生态环境数据建立模型进行冬小麦品质分区监测是可行的,且精度更高。本研究的创新点包括：1.引入非线性的神经网络研究冬小麦蛋白质含量与生态因子间的复杂关系,定量分析了影响冬小麦蛋白质含量关键的生态因子。2.构建了基于关键生态因子的权重的冬小麦品质分区模型,使品质分区更科学合理。3.提出了综合生态因子和遥感数据进行冬小麦品质遥感监测方法,并构建了综合模型,提高了冬小麦籽粒蛋白质含量监测精度。更多还原

【Abstract】 Ecological factors play an important role in affecting the grain quality of wheat. Ecological factors and remote sensing division are used to monitor the grain protein content of wheat through remote sensing data, in order to improve monitoring accuracy. Winter wheat in Beijing was taken as a case study; the main studies were as follows:(1)The preliminary screening analysis for key ecological factors of the grain protein content of winter wheat based on neural networks was carried out. For affecting the percentage of the protein content of winter wheat, the neural network algorithm was adopted to determine the influence of various factors on the grain quality of winter wheat. (2)Various factors and their corresponding weights were utilized to establish one division model, while these factors were only used to build another model. Based on the two division models, key ecological factors of the grain quality of winter wheat were used to select the optimization model to achieve the classification division of the grain quality of the high-quality winter wheat in Beijing. (3) The model of monitoring the grain quality of winter wheat was established by using ecological factors and remote sensing information for different divisions. It was compared with the remote sensing monitoring model and ecological factor model. The optimization model of monitoring the grain quality of winter wheat was adopted to monitor the grain quality of winter wheat in Beijing using remote sensing data. The study has a great significance in enhancing the remote sensing monitoring mechanism of the grain quality of winter wheat and improving monitoring accuracy.The main results were as follows:1. The preliminary analysis for key ecological factors of the grain protein content of winter wheat based on neural networks.The meteorological data and soil nutrient data of the most representative cultivation sites of winter wheat in Beijing were used to evaluate the relative importance of the grain protein content of winter wheat affected by temperature, precipitation, light and soil nutrient content, respectively based on the neural network method. The study showed that the main factors affecting the protein content of winter wheat in Beijing were the light duration in the dough stage (June 6th to June 10th), days of temperature above 32℃, the soil nitrogen content, the average temperature in the whole filling stage (early May to early June), the average temperature in the late filling stage (May 26th to May30th), the accumulated temperature above 0℃in the late filling stage(late May to early June), the average temperature in the milk stage(June 1 st to June 5th), the temperature in the late filling stage(late May to early June), the precipitation and soil organic matter content(late May to early June). Key ecological factors were adopted as the inputs of the neural network model. The model was used to produce a response curve to reflect the trend of the protein content.2. Study on the grain quality division of winter wheat in Beijing based on remote sensing and geographic information system.Under the support of ArcGIS, various key ecological factors of the protein content of winter wheat were used to make the spatial interpolation, and then the point-like key factors were interpolated to the raster and the multi-factors spatial database were established. Each factor was set a weight for its RATIO calculated by the neural network method. The meteorological factors and soil factors were analyzed in the ENVI environment, and then the division results derived from the equal weight and different weight models were compared. Different classifications of various divisions of the grain quality of winter wheat in Beijing were obtained; these results were analyzed.3. Remote sensing monitoring for the protein content in key periods of the grain quality of winter wheat based on divisions.The grain protein of winter wheat, Zhongyou 206 in Beijing was chosen as the research objective. Various vegetation indices (VIs) and the grain protein of zhongyou 206 in a variety of different growth stages of winter wheat in Beijing were calculated by using remote sensing data to study their correlation. The results showed that NDVIgreen in May 11th was best correlated with the grain protein, and the highly significance was achieved, thus, this period is the best phase for monitoring the grain quality of winter wheat using remote sensing. The ecological environmental data, spectral data were used to construct the spectral quality model, the ecological environmental quality model and the comprehensive model of spectral and ecological environmental quality. F test was used for these three models, and these models reached the highly significant level; Compared with the spectral quality model and the ecological environmental quality model, the coefficient of determination (R2) of the comprehensive model of spectral and ecological environmental quality had significantly improved, and the relative root mean square error (RRMSE) and relative error (RE) had sharply deceased. It was demonstrated that the comprehensive model of spectral and ecological environmental quality was obviously better than another two models. In the study of constructing model of the five divisions, the accuracy of sub-models had different degrees of increase compared with the whole model. The prediction accuracies in the most high-quality division, the second-class division, the third-class division, the fourth-class division and the worst quality division were 91.6%,89.3%,85.6% 83.6% and 92.2% respectively. In the comparison to the whole model, The predicting accuracies of the models constructed by using divisions had increased by 12%,10.5%,7.3%,11.7% and 14.3% respectively. Therefore, remote sensing and ecological environmental data were used to construct the model to estimate the grain quality of winter wheat under the different divisions. It was feasible and highly accurate.The innovations of this study include:1. the nonlinear neural network method is adopted to study the complex relationships between the protein content of winter wheat and ecological factors.2. The division model of the grain quality of winter wheat is constructed through key ecological factors and their weights. The quality division is more scientific and reasonable.3. The remote sensing monitoring method for the grain quality of winter wheat, combined with ecological factors and remote sensing data, is proposed, and an comprehensive model is constructed to improve the monitoring accuracy of the grain protein content of winter wheat.更多还原