A 3D spectral compensation method on close-range hyperspectral imagery of plant canopies

Abstract

Rapid and accurate plant phenotyping is vital to plant breeding and monitoring. Hyperspectral imaging (HSI) is the popular phenotypic technique to acquire spectral and spatial information of plants. However, the close-range HSI of plant canopies is greatly affected by the complex interaction of canopy geometry with illumination, which leads to biased or contaminated spectral information. Thus, the mitigation of these effects is imperative but challenging. In this study, a three-dimensional (3D) spectral compensation method on close-range canopy HSI was proposed, to correct the reflectance of canopies affected by imaging distance and leaf angle. First, the hyperspectral and depth images of canopies were registered and fused to generate hyperspectral 3D point clouds. Next, the full-spectrum reflectance on canopies was compensated based on the depth and angle information provided by the hyperspectral 3D point clouds. Then, the performance of spectral compensation results was evaluated by cluster analysis, spectral curve validation, and chlorophyll regression. Results on two plant types (perilla and tea seedlings) showed that after spectral compensation, the reflectance variations within canopies reduced greatly, the reflectance of whole canopies became more homogeneous, with a dominant cluster accounting for over 67% pixels of canopies. And using the mean reflectance curves of in vitro flattened leaves as the reference, the canopy reflectance after compensation were closer to the reference level, that the Euclidean Distance (ED) between them reduced by 50.6%. The determination coefficient (R²) for chlorophyll regression after compensation reached 0.75, increasing about 17% compared to that before compensation. The overall results demonstrated that the proposed 3D spectral compensation method was effective in mitigating the effects of imaging distance and leaf angle on plant canopies in close-range HSI. This could further facilitate the revelation of plant optical characteristics, which is of high significance for the accurate close-range plant phenotyping.

Plant phenotyping relevance

植物キャノピーの近接ハイパースペクトル画像に対する3Dスペクトル補償法を開発し、複数の評価で性能検証しており、表現型取得・抽出手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.