Temporal forecasting of plant height and canopy diameter from RGB images using a CNN-based regression model for ornamental pepper plants (Capsicum spp.) growing under high-temperature stress

Abstract

Abstract Being capable of accurately predicting morphological parameters of the plant weeks before achieving fruit maturation is of great importance in the production and selection of suitable ornamental pepper plants. The objective of this article is evaluating the feasibility and assessing the performance of CNN-based models using RGB images as input to forecast two morphological parameters: plant height and canopy diameter. To this end, four CNN-based models are proposed to predict these morphological parameters in four different scenarios: first, using as input a single image of the plant; second, using as input several images from different viewpoints of the plant acquired on the same date; third, using as input two images from two consecutive weeks; and fourth, using as input a set of images consisting of one image from each week up to the current date. The results show that it is possible to accurately predict both plant height and canopy diameter. The RMSE for a forecast performed 6 weeks in advance to the actual measurements was below 4.5 cm and 4.2 cm, respectively. When information from previous weeks is added to the model, better results can be achieved and as the prediction date gets closer to the assessment date the accuracy improves as well.

Plant phenotyping relevance

RGB画像からCNNで植物体高と樹冠径を予測し、予測精度を評価する手法開発・検証が研究の中心であるため。

Code and data availability

The paper's curated dataset of morphological measurements (plant height, canopy diameter) and weekly RGB photographs of the 15 Capsicum accessions is explicitly deposited on Zenodo with a DOI matching an allowed URL. No author analysis code or trained model checkpoints are stated as publicly available.