基于无人机遥感的玉米表型信息提取技术研究

发布时间：2018-11-19 18:12

【摘要】：表型信息是农作物品种、生长状况的直观表现,也是影响农作物产量的重要因素。随着全球人口基数不断增大,粮食需求量与日俱增,粮食供给问题日益严峻。快速精确提取大尺度农田中农作物的表型信息,监测作物的生长状况,并及时采取有效的管理措施,对选育高产优质的作物品种,维护我国粮食安全具有深远意义。然而,目前多采用人工实地测量的方法获取表型信息,虽准确性高,但区域覆盖度低,不适合大区域尺度育种试验田。随着遥感技术的飞速发展,使实时、快速、无损的获取大区域尺度的地表信息成为可能。本研究旨在为基于微小型无人机高通量遥感平台获取作物表型信息提供一定的理论依据,为研究玉米品种基因型与表型信息关联规律提供辅助支持。于2015年6月至9月在“国家精准农业示范研究基地”玉米育种研究区开展了关于微小型无人机高通量表型信息获取试验,进行了图像特征提取(玉米株高、抽雄时间、植被覆盖度、叶色变化)及LAI反演研究。主要研究工作及研究结果如下:(1)利用无人机高通量遥感平台获取的高清数码照片数据,采用ISODATA方法、SVM方法、基于HSV色彩空间变换的决策树分类三种方法进行冠层覆盖度提取,总精度和Kappa系数分别为59.06%、0.26,92.70%、0.96,98.32%、0.96。可见基于HSV色彩空间变换的决策树分类精度最高,可利用该方法进行多生育期影像冠层覆盖度提取。(2)利用基于HSV色彩空间变换的决策树分类和面向对象分类(结合纹理、HSV色彩空间变换、NDI植被指数、几何信息)两种方法进行玉米雄穗提取,分类总精度分别为83.79%、85.91%,相比于基于HSV色彩空间变换的决策树分类方法,面向对象分类方法精度较高。因此,利用面向对象分类方法提取玉米雄穗,进而提取玉米的抽雄时间,提取精度为65.62%,可见利用该方法进行玉米抽雄时间的提取、监测是可行的。(3)利用基于HSV色彩空间变换的决策树分类进行多生育期玉米叶色变化提取。利用影像的色调值可以显著区分叶片的颜色,从而达到提取玉米叶色的需求。(4)利用多光谱影像提取的8种植被指数反演LAI时,对单变量模型而言,NDVI反演的多生育期LAI效果好于其他植被指数,其中线性模型和幂模型反演结果的R2和RMSE分别为0.525、0.711,0.530、0.717,可以用NDVI来监测多生育期玉米LAI的变化情况;在多变量反演过程中,首先对8种植被指数进行主成分分析得到主成分变量,然后对主成分变量进行多元线性回归和BP神经网络分析。结果显示,BP神经网络对玉米LAI具有较好的反演能力,R2为0.608,RMSE为0.745,可以较好的预测多生育期玉米LAI的变化情况。(5)在提取株高的过程中,DSM影像提取的6884份育种材料株高值与实测株高值具有很好的线性关系,R2为0.527,RMSE为0.223。因此通过该方法可以代替传统的人工田间测量株高方式;通过生成的株高分布图,可以更直观的看到株高的分布及变化情况等信息。
[Abstract]:Phenotypic information is the visual expression of crop variety and growth condition, and it is also an important factor affecting crop yield. With the global population base increasing, food demand is increasing, food supply problem is becoming more and more serious. Rapid and accurate extraction of phenotypic information of crops in large-scale farmland, monitoring of crop growth, and timely and effective management measures are of far-reaching significance for breeding high yield and high quality crop varieties and maintaining food security in China. However, at present, artificial field measurements are used to obtain phenotypic information. Although the accuracy is high, but the area coverage is low, it is not suitable for large-scale breeding field. With the rapid development of remote sensing technology, it is possible to obtain large scale surface information in real time, fast and lossless. The purpose of this study is to provide a theoretical basis for obtaining crop phenotypic information based on the micro-UAV high-throughput remote sensing platform, and to provide auxiliary support for studying the correlation between genotype and phenotypic information of maize varieties. From June to September 2015, an experiment on obtaining high-throughput phenotypic information of micro-UAV was carried out in the Maize breeding Research area of the National Precision Agriculture demonstration Research Base. Vegetation coverage, leaf color change) and LAI inversion. The main research work and results are as follows: (1) using high-throughput remote sensing platform of UAV to obtain high-definition digital photo data, using ISODATA method, SVM method, Three methods of decision tree classification based on HSV color space transform were used to extract canopy coverage. The total accuracy and Kappa coefficient were 59.06 and 0.2692.70 respectively. It can be seen that the classification accuracy of decision tree based on HSV color space transformation is the highest, and this method can be used to extract canopy coverage of multi-growth image. (2) decision tree classification based on HSV color space transformation and object oriented classification (combined with texture). HSV color space transformation, NDI vegetation index and geometric information are used to extract maize male ear. The total classification accuracy is 83.79 and 85.91, respectively, compared with the decision tree classification method based on HSV color space transformation. The accuracy of object-oriented classification method is high. Therefore, using the object oriented classification method to extract the male ear of maize, and then to extract the heading time of maize, the extraction accuracy is 65.622. It can be seen that this method is used to extract the heading time of maize. Monitoring is feasible. (3) the decision tree classification based on HSV color space transformation is used to extract the color change of maize leaves in multi-growth period. The color of leaves can be distinguished significantly by using the hue value of image. (4) when LAI is retrieved exponentially from 8 planting plants extracted from multispectral images, for the single variable model, the color of maize leaves can be extracted. The effect of LAI inversion by NDVI was better than that of other vegetation indices. The R2 and RMSE of linear model and power model were 0.5250.70110.530300.717, respectively. NDVI could be used to monitor the change of LAI in maize in multi-growth period. In the process of multivariate inversion, the principal component variables are obtained by principal component analysis (PCA) for the 8-implant index, and then the principal component variables are analyzed by multivariate linear regression and BP neural network. The results showed that BP neural network had a better ability to retrieve maize LAI (R2 = 0.608 RMSE = 0.745), which could be used to predict the variation of LAI in maize at multi-growth stage. (5) in the process of extracting plant height, There was a good linear relationship between the plant height of 6884 breeding materials extracted by DSM image and the measured plant height. The R2 was 0.527 and the RMSE was 0.223. Therefore, this method can replace the traditional way of measuring plant height in artificial field, and the information of plant height distribution and variation can be seen more intuitively by generating plant height distribution map.
【学位授予单位】：东北农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S513;TP751

【参考文献】