Digital Phenotyping in Plant Breeding: Evaluating Relative Maturity, Stand Count, and Plant Height in Dry Beans via RGB Drone-Based Imagery and Deep Learning Approaches

Abstract

The datasets comprised were collected from the MSU dry bean breeding program in 2022 planting season at SVREC location. The dataset includes the orthomosaic (.tif) generated using the raw images, raw images (.jpeg) from two flight altitude, plot boundary delimitations (.shp), clipped plots, annotations, Faster R-CNN deep learning model, and the ground-truth (GT) notes to predict stand count (SC). /a._Orthomosaics: This directory contains 1 file 1. Image naming structure: `Date-of-flight _ _ /` Orthomosaic (TIFF image) collected around VC growth stage containing readable EXIF headers with image metadata. /b._Shapefiles Plot boundaries files (.shp) and field area from 2022 SVREC location containing plot level information using the breeding program metadata. /c._ClipPlots Image naming structure: `Experiment-PlotID _ _ ` PNG images clipped from the orthomosaic of 2022 SVREC location. /d._Annotations__1 VGG project containing the bean plant annotations via boundary boxes with x, y, height and width coordinates. /e._Resize_img_annot__2 VGG project containing the bean plant annotations via boundary boxes with x, y, height and width coordinates. /f._HyperTunning__3 Model used to perform the hyperparameter tunning. /g._PseudoLab__4: Pseudo labeling using repetitions 3 and 4 from the 2022 SVREC location. This folder contains a subfolder: Image naming structure: `Experiment-PlotID _ _ ` PNG images clipped from the orthomosaic of rep 2 and 4 of the 2022 SVREC location. /h._TrainModel__5 Model used to perform the training. /i._TestingModel1: This folder contains orthomosaic (.tif), shapefile (.shp), clipped plots (.png) and the inference model to early flight date. /i._TestingModel2 This folder contains orthomosaic (.tif), shapefile (.shp), clipped plots (.png), new set of annotations, and the inference model to lower flight altitude (6 meters). /i._TestingModel3 This folder contains orthomosaic (.tif), shapefile (.shp), clipped plots (.png) and the inference model to higher flight altitude (10 meters). /j._Mask_count_Seg-CNN Traditional methods using mask via R and Python programing to perform stand count (SC). Also, the CNN segmented model is available to classify between soil and vegetation. /Raw_img3_6_13_22_SVREC_RGB_SC_7m_1: Raw images (.jpeg) collected at 7 meters of flight altitude. /Raw_img3_6_13_22_SVREC_RGB_SC_7m_2: Cont. Raw images (.jpeg) collected at 7 meters of flight altitude. /Raw_img4_6_13_22_SVREC_RGB_SC_10m_1: Raw images (.jpeg) collected at 10 meters of flight altitude. /Raw_img4_6_13_22_SVREC_RGB_SC_10m_2: Cont. Raw images (.jpeg) collected at 10 meters of flight altitude.

Plant phenotyping relevance

RGBドローン画像と深層学習・セグメンテーションにより、乾燥豆のスタンド数、成熟度、草丈を推定するデータセットおよび解析手法が中心である。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.