Monitoring cotton (Gossypium hirsutum L.) germination using ultrahigh-resolution UAS images
Monitoring cotton (Gossypium hirsutum L.) germination using ultrahigh-resolution UAS images
复制标题
DOI：
10.1007/s11119-017-9508-7
复制
发表时间：
2018-02-01
影响因子：
6.2
通讯作者：
Maeda, Murilo M.
中科院分区：
文献类型：
--
作者：
Chen, Ruizhi;Chu, Tianxing;Maeda, Murilo M.
种子发芽率的检查对于种植者在季节早期确定是否需要重新种植田地非常重要。本研究的目的是探索利用无人机系统（UAS）的可见光波段图像监测和量化棉花发芽过程的潜力。使用了一个轻型UAS平台，它携带了一个消费级的红色，绿色和蓝色相机，由内置的万向节系统稳定。为了在发芽阶段获得高分辨率图像，无人机系统平台在离地面约15-20米的高度飞行。通过应用运动恢复结构（SfM）算法，图像被校正并且以大约6-9 mm/像素的地面采样距离进行正射镶嵌。提出了一种新的解决方案，用于计算平均植物的大小和发芽棉花植株的数量，根据提取的叶片多边形从正射影像。通过使用估计的发芽棉花植物的数量，植物密度和累积发芽率也可以以直接的方式使用特定于田地的参数来估计。通过比较从六个棉花行段收集的地面观测数据对发芽棉花植株的估计数量进行了评估。结果表明，该方法对棉花发芽株数的平均估计准确率达到88.6%。如果采用具有近红外波段的图像，则可以进一步提高精度。
Examination of seed germination rate is of great importance for growers early in the season to determine the necessity for replanting their fields. The objective of this study was to explore the potential of using unmanned aircraft system (UAS)-based visible-band images to monitor and quantify the cotton germination process. A light-weight UAS platform was used, which carried a consumer-grade red, green, and blue camera stabilized by a built-in gimbal system. In order to obtain ultrahigh image resolution during the germination stage, the UAS platform was flown at an altitude of approximately 15-20 m above ground. By applying the structure-from-motion (SfM) algorithm, the images were rectified and orthographically mosaicked with a ground sampling distance of approximately 6-9 mm/pixel. A novel solution was then developed for calculating the average plant size and the number of germinated cotton plants according to the leaf polygons extracted from the orthomosaic images. By using the estimated number of germinated cotton plants, the plant density and the cumulative germination rate can also be estimated in a straightforward manner using field-specific parameters. An assessment of the proposed solution was conducted by comparing the estimated number of the germinated cotton plants against ground observation data collected from six cotton row segments. The results demonstrated that the average estimation accuracy achieved 88.6% in terms of identifying the number of the germinated cotton plants. The accuracy may be further improved if images with near infrared band are employed.