基于无人机高清数码影像的水稻产量估算

发布时间：2018-03-16 03:14

本文选题：无人机　切入点：颜色空间　出处：《沈阳农业大学学报》2017年05期 　论文类型：期刊论文

【摘要】：目前常用的水稻产量估算方法以卫星遥感估产为主,卫星遥感估产的分辨率较低、缺乏机理性、误差较大。为了能够快速灵活地获取水稻冠层信息、提高分辨率、准确地估测水稻产量,利用无人机平台搭载高清数码相机,拍摄从抽穗期到成熟期的水稻冠层影像,首先应用中值滤波算法对RGB颜色空间下水稻冠层图像进行去噪,然后针对彩色水稻图像的颜色特征,将图像由RGB颜色空间转换到L*a*b*颜色空间,运用K均值聚类算法对水稻冠层图像进行聚类分析、图像分割,提取出水稻穗、获得水稻穗数量、代入水稻产量估算公式进行估产。试验区域共有18块水稻小区(长8m,宽5m),在水稻抽穗期到成熟期之间拍摄4次。试验记录的数据包括拍摄的时间、高度以及分辨率,同时还要在田间实测水稻穗的数量和水稻的产量,为后期评价和判断K均值聚类算法提取水稻穗的精度以及水稻产量估测的精度提供依据。对水稻产量的实测值与估测值、田间实测的水稻穗数量与图像中提取水稻穗数量进行对比分析。结果表明:对8月18日无人机拍摄的水稻冠层影像进行图像分割,提取出水稻穗的效果较好,估产的精度较高,产量估计均方根误差和平均绝对百分误差分别为9.08和22.8%,水稻穗数估计均方根误差和平均绝对百分误差分别为19.86和5.8%。说明利用无人机搭载数码相机能够快速、无损地获取水稻冠层信息,运用K均值聚类算法能够较为准确地将水稻穗从水稻冠层图像中分割出来,利用数字图像对水稻产量进行估算是可行的。
[Abstract]:In order to obtain rice canopy information quickly and flexibly and improve the resolution, the commonly used methods for estimating rice yield are satellite remote sensing, which has the advantages of low resolution, lack of mechanical rationality and large error. The rice yield was estimated accurately, and the high-definition digital camera was used on the UAV platform to capture the rice canopy image from heading stage to mature stage. Firstly, the median filtering algorithm was used to de-noise the rice canopy image in RGB color space. Then according to the color characteristics of the color rice image, the image is transformed from RGB color space to Lena color space. The K-means clustering algorithm is used to cluster the rice canopy image, image segmentation, rice panicle extraction, and the number of rice panicles is obtained. A total of 18 rice plots (8 m in length and 5 m wide) were used to estimate rice yield. The data recorded in the experiment included the time, height and resolution of the shoot, which were taken 4 times between the heading stage and the maturity stage of rice. At the same time, the number of rice panicles and the yield of rice were measured in the field, which provided the basis for the later evaluation and judgement of the precision of rice panicle extraction by K-means clustering algorithm and the precision of rice yield estimation. The number of rice panicles measured in the field was compared with the number of rice panicles extracted from the images. The results showed that the rice crown images taken by UAV on August 18th had a good effect of extracting rice panicles and the precision of estimating yield was higher. The root mean square error and average absolute percent error of yield estimation are 9.08 and 22.80.The root mean square error and average absolute percentage error of rice panicle estimation are 19.86 and 5.80.It shows that using UAV to carry digital camera can be rapid. Using K-means clustering algorithm, rice panicles can be segmented accurately from rice canopy images, and it is feasible to estimate rice yield by using digital images.
【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】：国家重点研发项目(2016YFD0200700,2017YFD0300706) 辽宁省教育厅课题重点项目(LSNZD201605)
【分类号】：S127;S511

【相似文献】