几种经济作物的光谱识别研究及应用

发布时间：2018-04-06 20:38

本文选题：实测光谱　切入点：光谱识别　出处：《新疆大学》2014年硕士论文

【摘要】：21世纪人类将步入信息社会，要求农业由传统向现代的转变，经营方式由粗放型向集约型转变。当前，我国正面临着越来越尖锐的资源与环境问题，为保障13亿人口的食物安全，推动农业科学技术的发展、实施可持续发展战略势在必行。地界学的专家也认识到“数字地球战略”将是推动我国信息化建设和社会经济、资源环境可持续发展的重要武器，发展现代农业始终成为农业可持续发展的必由之路。农作物类型的识别和作物某些参数(如叶绿素、叶面积指数等)及产量的估测在农业生产中非常重要；由于人类环境在作物生长过程中的影响越来越大，适时快速检测作物类型及作物叶绿素含量显得愈加重要；高光谱技术通过对作物光谱反射率及其与叶绿素含量关系的研究分析已经成为作物监测及估测的强有力工具，并且其也是现代农业研究的一个方面。经济作物生产是高效益型的农业经济产业，一个区域的经济作物生产绝大部分参与市场流通，农民衣住行的费用都是从经济作物的收益中获取。自党的十一届三中全会提出的“决不放松粮食生产，积极发展多种经营”方针以来，各地区都对农业结构进行调整，其中，经济作物所占比例有所增大，农民收入及农业生产也得到提高与发展。在工业化、城镇化深入发展中同步推进农业现代化，发展现代农业，提高农业发展的科技含量，是十二五时期农业生产的一项重大任务。因此，利用光谱技术对经济作物进行研究具有实际意义。本论文利用美国ASD（analytical spectral device)公司的ASD FieldSpecPro光谱仪和SPAD·502便携式叶绿素仪于2012年5-9月测量了研究区三种经济作物的高光谱反射率与叶绿素含量，分析了打瓜、甜菜和葫芦的光谱特征；采用基于主成分分析的BP神经网络技术及主成分分析的Fisher判别模型对三种作物进行光谱识别研究并进行对比；基于特征光谱及植被指数对叶绿素含量进行了预估，并对作物进行了简单估产研究。得到了以下研究结果： 1、将作物实测光谱数据进行预处理，然后利用主成分分析对其进行降维处理，提取到5个主成分，其累计贡献率达99.79%，能很好解释原始光谱的全部信息；以5个主成分得分值为输入变量，作物种类为输出变量，建立PCA-BP模型和PCA-FDA模型，前者拟合残差为5.2156×10-6，对15个未知样本的识别率达到100%，后者待判样本回带验证率及检验样本识别率均达100%；PCA-BP模型是通过不断调整隐含层的节点数来优化模型结构以及其人为设定预测结果中作物区分的界限偏差，使得受主观因素影响较大，而PCA-FDA模型操作简单明了且客观性强、识别结果更科学。 2、打瓜叶绿素含量与所选光谱指数相关性均达到极显著水平，其中以MSAVI2与打瓜叶绿素含量相关性(R2=0.82)最好；且以其为自变量，叶绿素含量为因变量建立的三项式函数模型： Y=393.91x3-580.68x2+327.49x-4.1198精度最高（R2=0.8066），成为打瓜叶绿素含量最佳估算模型。葫芦叶绿素含量估算中，基于RNDVI的叶绿素估算模型的确定性系数R2为0.7376，RMSE为1.0249，优于其他植被指数模型；较于前者，基于主成分的对数模型预测值确定性系数R2为0.8201，RMSE为0.9126，为最佳叶绿素估算模型。 3选取的植被指数中，甜菜产量与VARI在各生育期相关性最好，均达到显著性水平，相关系数最大的为块根膨大期（0.8306），与其次为叶丛繁茂期（0.8107），苗期（0.8076），，糖分累积期（0.8015）；VARI与甜菜产量在4个生育期建立的回归方程相关系数均达到极显著相关水平，块根膨大期一元三次回归方程RMSE最小为0.0882，单时相估产精度均高；四个生育期复合回归模型检验R2与RMSE分别为0.960和0.127，精度最高，估算效果最好。
[Abstract]:In twenty-first Century, mankind will enter the information society, the requirement of agricultural transformation from traditional to modern, changing the operating mode from extensive to intensive. At present, our country is facing the problems of environment and resources is more and more sharp, 1 billion 300 million of the population for the protection of food safety, promote the development of agricultural science and technology, it is imperative to implement the strategy of sustainable development. Border studies are also aware of the "Digital Earth" strategy will be to promote the informationization construction and the social economy in our country, an important weapon for the sustainable development of resources, the development of modern agriculture has become the only way which must be passed to sustainable agricultural development.
Some parameters identification and crop crop types (such as chlorophyll, leaf area index) and yield estimation is very important in agricultural production; due to human environmental impact in the crop growth process more and more timely, rapid detection of crop types and crop chlorophyll content becomes more and more important; hyperspectral technology has become a powerful and crop monitoring the estimation tool through the research and Analysis on crop spectral reflectance and its relationship with chlorophyll content, and is also one aspect of the modern agricultural research.
Economic crop production is high benefit agricultural economy industry, production of a regional economic crop in the vast majority of farmers participate in the market circulation, housing bank charges are obtained from the economic crop income. Since the party's proposed the third Plenary Session of the 11th CPC Central Committee "no grain production, has been actively developing a diversified economy" policy, by region all the adjustment of agricultural structure, the economic crop proportion increased, the income of the farmers and agricultural production has been improved with the development in industrialization, urbanization development in promoting the modernization of agriculture, the development of modern agriculture, improving agricultural science and technology development, is an important task for agricultural production in 12th Five-Year so period. The study of economic crops, which has practical significance to use spectral techniques.
In this paper, using the ASD (analytical spectral device), ASD FieldSpecPro and SPAD 502 spectrometer portable chlorophyll meter measuring the three economic crops in the study area of high spectral reflectance and chlorophyll content in 2012 5-9 months, analyzed the spectral characteristics of melon, sugar beet and gourd; the study on the evaluation model of spectral recognition on three kinds of crops and compared the BP neural network technology of principal component analysis and principal component based on Fisher; spectral characteristics and vegetation index were estimated based on the content of chlorophyll, and the crops were a simple estimation research. The following results were obtained:
1, the crop measured spectral data preprocessing, then use principal component analysis to reduce the dimension of the extracted 5 main components, the cumulative contribution rate of 99.79%, can well explain all information of the original spectrum; the 5 principal component score values as input variables, output variables for crop types the establishment of PCA-BP model, and PCA-FDA model, the fitting error is 5.2156 * 10-6, the identification of 15 unknown samples reached 100%, the latter to be sentenced to the sample with verification test sample rate and recognition rate reached 100%; the PCA-BP model is through continuous adjustment and optimization of model structure and its prediction result set in crop distinction the number of hidden layer nodes deviation, which is affected by the subjective factors, while the PCA-FDA model is simple and has strong objectivity, the recognition result is more scientific.
2, the chlorophyll content of watermelon and the selected spectral index correlation reached significant level, the correlation between MSAVI2 and chlorophyll content of Watermelon (R2=0.82) and its best; as independent variables, the chlorophyll content for three type variables to establish function model: Y=393.91x3-580.68x2 +327.49x-4.1198 (R2=0.8066), the highest precision become the best models for the estimation of chlorophyll content of watermelon estimation of chlorophyll content. RNDVI gourd, chlorophyll estimation model of the deterministic coefficient of R2 is based on 0.7376, RMSE was 1.0249, better than other vegetation index model; compared with the former, the prediction model of principal component values of the logarithmic deterministic coefficient R2 is based on the 0.8201, RMSE is 0.9126, the best for chlorophyll estimation model.
3 selection of vegetation index, yield and VARI of sugar beet best in different growth period are significant correlation, correlation coefficient, the maximum period of tuber enlargement (0.8306), and then leaves lush growing stage (0.8107), (0.8076), seedling sugar accumulation period (0.8015); the correlation coefficients of regression equation and VARI beet yield in the 4 stages of the establishment have reached a significant level, the period of tuber enlargement, one of the three regression equations of RMSE minimum is 0.0882, while the yield estimation accuracy is higher; the four stage composite regression model verification of R2 and RMSE were 0.960 and 0.127, the highest precision, the best estimate.

【学位授予单位】：新疆大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：O433;S31

【参考文献】