Automatic Sugar Beet Phenotyping in Open Field by a Computer Vision System

Abstract

Crop growth is an important quality assessment in plant breeding, especially in open field crops which grow in fluctuating and unfavorable outdoor conditions. To evaluate the growth potential of different plant varieties, researchers conduct leaf area measurements of emerged plants to evaluate its growth potential. This is a time consuming and labor intensive activity and therefore often only conducted on random spots on the field. An automatic computer vision system was built to automate and to speed up this plant phenotyping process. The system consist of three color cameras mounted on an implement facing straight downwards, lamps for illumination, an encoder wheel and a computer system. Natural light was blocked by a surrounding cover to limit the effect of variable outdoor light conditions on the image quality. The computer vision software makes use of an excessive green algorithm (2G - R - B) to segment the plant material from the soil. As the crop plants are sown by a precision sowing device in a regular pattern a method based on the fast-fourier transform (FFT) is used to distinguish crop plants from weed plants. A rectangular based clustering algorithm, based on 8-pixel nearest-neighbor connectivity, is used to cluster separated plant-parts together as one individual plant object used to measure the leaf area. The system was validated in an open-field sugar beet crop at the growing stage off our leaves. Fifty-five sugar beet plants were manually measured by experienced plant scouts(“ground truth”). The same plants were measured with the computer vision system. An ANOVA F-test(P<0.05) was used to discriminate the two measurement methods. The F-probability was 0.055 an djust above the significance level. So the H0 hypothesis that there is not a difference between human measurement and machine vision measurement was no trejected. Possible causes of difference was the inability of the system to detect and measure plants damaged by animals and very small plants which were occluded by clods or bigger plants. Nevertheless,with improvements on the vision software and camera/lamp configuration, the system is profitable for a fast and accurate leaf area measurement and corresponding plant phenotyping.

Plant phenotyping relevance

圃場でのビート個体の葉面積を自動取得するコンピュータビジョンシステムを開発し、手動測定と比較検証しており、植物表現型取得法が研究の中心である。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.