High-throughput Morphological Phenotyping of Tomato Seedlings Using Computer Vision and Machine Learning

Abstract

Monitoring the early-stage vegetative growth is particularly critical for determining the plant's vigour, stress tolerance, and yield potential. Accurate and timely phenotyping of these traits is essential for informed decision-making in cultivation. However, traditional phenotyping methods are labour-intensive and time-consuming since most of them rely on manual inspection and subjective criteria. This study presents an automated, image-based approach integrated with artificial intelligence to monitor morphological traits of early growth stages of tomato plants. In this study, images of the seedlings of the tomato variety called “Thilina” were used, and an Excess Green (ExG) based segmentation method was employed for image preprocessing. The K-nearest neighbour machine learning model has gained the highest accuracy of 89% out of all machine learning models, and the convolutional neural network was utilised in identifying the plant timelines with prominent growth. The results of seedling evaluation through image-based phenotyping reveal that early vegetative parameters are reliable predictors of transplant readiness and vigour.

Code and data availability

The article describes a tomato seedling image dataset (1920 images, 80 plants, 11 days) and ML models (k-NN, CNN), but contains no data availability statement, no public repository deposit, and no code/model release. No paper-specific public asset is available.

No evidence-backed public reproduction asset is currently recorded.