Prediction of germination ability of tomato seeds based on phenotypic traits using machine learning

Abstract

This research proposes a method to automatically detect the phenotypic traits of tomato seeds and predict the germinating ability before planting. The proposed autonomous screening method combines computer vision, and machine learning algorithms to provide real-time, non-invasive phenotypic analyses of tomato seeds. Seeds of tomato verity called “Thilina” which is one of the popular commercially grown tomato varieties in Sri Lanka were used for this experiment. An experimental setup with a camera was used for the acquisition of images of the seeds. The images were preprocessed and morphological traits were collected in every seed. Different machine learning models are trained and tested. Identification of the most effective model for predicting seed germination is determined based on the evaluation parameters of precision, recall, F1-score and accuracy. It is concluded that KNN is the best classifier where 94% of the test dataset is accurately predicted by the KNN model. The highlights of this study will be crucial to overcoming the labor-intensive nature and errors made when conducting conventional germination tests. To the best of our knowledge, this study is the first in Sri Lanka which automated conventional seed tests to predict the germination ability.