High throughput phenotyping of cotton plant height using depth images under field conditions

Plant height is an important phenotypic trait that can be used not only as an indicator of overall plant growth but also a parameter to calculate advanced traits such as biomass and yield. Currently, cotton plant height is primarily measured manually, which is laborious and has become a bottleneck for cotton research and breeding programs. The goal of this research was to develop and evaluate a high throughput phenotyping (HTP) system using depth images for measuring cotton plant height under field conditions. For this purpose, a Kinect-v2 camera was evaluated in a static configuration to obtain a performance baseline and in a dynamic configuration to measure plant height in the field. In the static configuration, the camera was mounted on a partially covered wooden frame and oriented towards nadir to acquire depth images of potted cotton plants. Regions of interest of plants were manually selected in the depth images to calculate plant height. In the dynamic configuration, the Kinect-v2 camera was installed inside a partially covered metal-frame that was attached to a high-clearance tractor equipped with real time kinematic GPS. A six-step algorithm was developed to measure the maximum and average heights of individual plots by using the depth images acquired by the system. System performance was evaluated on 108 plots of cotton plants. Results showed that the Kinect-v2 camera could acquire valid depth images of cotton plants under field conditions, when a shaded environment was provided. The plot maximum and average heights calculated by the proposed algorithm were strongly correlated (adjusted R2 = 0.922–0.987) with those measured manually with accuracies of over 92%. The average processing time was 0.01 s to calculate the heights of a plot that typically has 34 depth images, indicating that the proposed algorithm was computationally efficient. Therefore, these results confirmed the ability of the HTP system with depth images to measure cotton plant height under field conditions accurately and rapidly. Furthermore, the imaging-based system has great potential for measuring more complicated geometric traits of plants, which can significantly advance field-based HTP system development in general.

Detailed information refers to our paper Jiang, Y., C. Li., and A. Paterson. 2016. High-throughput phenotyping of cotton plant height using depth images under field conditions. Computers and Electronics in Agriculture. 130 (2016): 57-68.

figure2

 

Figure 1. Diagram of the ground-based data acquisition system.

figure6

Figure 2. Representative results of (A) the acquired depth images retrieved from binary files, (B) the stitched depth images for plots with various germination conditions, and (C) an example of spatial discontinuities in the stitched depth images. The intensity represents the depth measurement of a pixel: pixels of higher plants are hypo-intense whereas those of lower plants are hyper-intense in the depth images. In panel (A), the red rectangles are the sub-images selected for depth image stitching.

figure7

Figure 3. Results of linear regression analyses between Kinect-v2 and manual measurements of plot maximum height for (a) all 108 plots and (b) 72 plots that contained no plant taller than 1350 mm.

figure8

Figure 4. Results of linear regression analyses between Kinect-v2 and manual measurements of plot average height for (a) all 108 plots and (b) 72 plots that contained no plant taller than 1350 mm.