Drone Prediction of Individual Chinese Cabbage Weight | 53 Days Before Harvest

A 2023 Scientific Reports study moved UAV crop prediction from field-level yield toward individual-plant harvest weight. Researchers monitored 1,196 Chinese cabbage plants with RGB and multispectral cameras mounted on UAVs.

Table of Contents

RGB orthomosaics detected more than 95% of plants
Multi-temporal features predicted harvest weight with R² = 0.86
Useful predictions were possible 53 days before harvest
The model is not automatically portable to every cabbage field
Reference

RGB orthomosaics detected more than 95% of plants

Drone imaging of a Chinese cabbage field — RGB and multispectral UAV data were collected across multiple dates.

An object-detection pipeline identified more than 95% of individual plants from RGB orthomosaic imagery. Tracking the same plant across dates allowed the researchers to build multi-temporal morphological and spectral features.

Multi-temporal features predicted harvest weight with R² = 0.86

Automatic detection of individual Chinese cabbage plants — Plant-level features were tracked through time and linked to final harvest weight.

The best model achieved R² = 0.86 and RMSE = 436 g per plant. The temporal sequence mattered: combining features from multiple growth stages improved prediction compared with treating the field as a single snapshot.

Useful predictions were possible 53 days before harvest

Even up to 53 days before harvest, predictions remained above R² = 0.72 with RMSE below 560 g per plant. This creates potential value for harvest planning and high-throughput phenotyping.

The model is not automatically portable to every cabbage field

Workflow for UAV-based cabbage weight prediction — The workflow combines individual detection, feature extraction and multi-temporal modeling.

The experiment used 1,196 plants under a particular experimental-field design. Cultivar, season, region, soil, sensor, flight conditions and management can all affect model transferability. Individual predictions also retain errors of several hundred grams.

The main contribution is therefore a workflow for turning UAV imagery into longitudinal plant-level data, not a universal ready-made cabbage scale.

Reference

Aguilar-Ariza A et al. UAV-based individual Chinese cabbage weight prediction using multi-temporal data. Scientific Reports. 2023;13:20122. https://doi.org/10.1038/s41598-023-47431-y

Predicting Individual Chinese Cabbage Weight by Drone: R² > 0.72 Even 53 Days Before Harvest