Close-range multispectral imaging with Multispectral-Depth (MS-D) system

Abstract

In this work, a Multispectral-Depth (MS-D) imaging system for close range plant inspection is presented. The proposed system is comprised of multispectral cameras calibrated with respect to an RGB-D camera. It is ultimately tested and quantitatively compared to state-of-the-art feature-based methods for multispectral image registration. The results show that MS-D system outperforms state-of-the-art methods in all experimental trials, from registration of checkerboard images (where the accuracy of the MS-D system was on a sub-pixel level) to registration of feature-rich plant images, both real and synthetic. While the greatest registration error of the MS-D system amounted to 9 pixels, registration error of the feature-based method was up to 19 times greater. Additionally, contrast to the state-of-the-art feature matching approaches, the MS-D system, once calibrated, is applicable as is, without the need for recalibration. As a part of this work, MS-D has been deployed in encapsulated growth chambers for rapid data collection and in a small indoor organic farm for automated plant monitoring. As a part of the experimental plant monitoring study, it was shown that vegetation indices calculated with the MS-D system can be used to estimate water stress in Spathiphyllum plants equally well as with the spectroradiometer, human operated device for measurement of plant vegetation indices. The biggest relative change between the vegetation indices calculated for the plants exposed to the short term water stress and the control group was found in values of NDRE, amounting to 94.8%, followed by the relative change of 64.9% observed for the values of SR.

Plant phenotyping relevance

MS-Dマルチスペクトル・深度画像システムを開発し、画像レジストレーション性能を定量比較するとともに、植物モニタリングと水ストレス推定へ応用しており、植物表現型取得法が中心的である。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.