东北粳稻叶绿素相对含量的无人机高清影像检测方法

发布时间：2018-05-02 16:18

本文选题：无人机 + 回归分析　；参考：《沈阳农业大学学报》2017年06期

【摘要】：叶绿素相对含量(soil and plant analyzer development,SPAD)是评价水稻健康状况的重要农学参数,为了解决传统监测方法工作量大,效率低的问题,以东北粳稻为研究对象,采用不同施肥处理开展小区试验,利用无人机低空遥感技术分别获取水稻分蘖期、拔节孕穗期、抽穗灌浆期水稻冠层高清数码影像,同时利用叶绿素仪测量水稻冠层SPAD值,并对无人机高清数码影像反演SPAD的可行性及方法进行研究。结合k-means聚类和阈值分割的方法去除背景提取出水稻叶片的RGB值,构建出R、G、B及G/R、G/B、B/R、R-B、G-R、NRI、NGI、NBI共11种颜色参数,并分别用11种参数和水稻叶片SPAD做相关性分析,分析结果表明NRI、B/R、R-B 3种参数和SPAD值高度相关。分别采用一元线性回归分析法和BP神经网络法对3种参数和SPAD的关系进行建模并对建模精度进行分析。结果表明:无人机高清影像反演SPAD是可行的,其中一元线性回归分析中,NRI和SPAD的建模精度高于B/R和R-B,均方根误差(RMSE)为1.51;基于NRI、B/R和R-B的多特征输入的BP神经网络预测粳稻SPAD的RMSE为1.354,相比基于NRI的一元线性回归分析模型精度提升11%,BP模型能较好地对东北粳稻的SPAD进行反演,能为无人机低空遥感反演粳稻SPAD提供理论依据和实现方法。
[Abstract]:Relative chlorophyll content (and plant analyzer) is an important agronomic parameter to evaluate rice health status. In order to solve the problem of heavy workload and low efficiency of traditional monitoring methods, Northeast japonica rice was used as the research object and different fertilization treatments were used to carry out plot experiment. The high-definition digital images of rice canopy at tillering stage, jointing and booting stage and heading and filling stage were obtained by using UAV low-altitude remote sensing technology, and SPAD values of rice canopy were measured by chlorophyll meter. The feasibility and method of SPAD inversion of UAV high-definition digital image are studied. In combination with k-means clustering and threshold segmentation, RGB values of rice leaves were extracted by removing background, and 11 color parameters were constructed, and 11 color parameters were analyzed by using 11 parameters and SPAD of rice leaves. The results show that there is a high correlation between the three parameters and the SPAD value. The relationship between the three parameters and SPAD is modeled by linear regression analysis and BP neural network, respectively, and the modeling accuracy is analyzed. The results show that the SPAD inversion of UAV high-definition image is feasible. The modeling accuracy of NRI and SPAD in univariate linear regression analysis is higher than that of B / R and R-Band the root mean square error (RMSE) is 1.51.The BP neural network based on NRI-B / R and R-B for predicting SPAD of japonica rice is 1.354, which is higher than that based on NRI. The SPAD of japonica rice in Northeast China can be retrieved by using BP model with improved accuracy of the model. It can provide theoretical basis and implementation method for retrieving japonica rice SPAD from UAV low altitude remote sensing.
【作者单位】：沈阳农业大学信息与电气工程学院/辽宁省农业信息化工程技术中心;
【基金】：国家重点研发计划项目(2016YFD020060307)
【分类号】：S511.22;TP751

【相似文献】