Variety Classification of Lactuca Sativa Seeds Using Single-Kernel RGB Images and Spectro-Textural-Morphological Feature-Based Machine Learning

Abstract

Growing lettuce become popular now and the use of specific seeds on a constraint environment relies on the proper phenotypic classification of seed germplasm. Lettuce cultivars are usually differentiated based on leaf characteristics when it is matured because its seeds are characterized by almost the same spectro–textural–morphological signatures. Visual inspection of small lettuce seeds leads to the subjective classification that is unideal for seed phenotyping. To overcome this agro–industrial challenge, computer vision was incorporated with computational intelligence. In this study, two types of Lactuca Sativa L. cultivars were used, namely grand rapid and Chinese loose–leaf lettuce seeds. A consumer–grade Huawei Nova 5T mobile phone camera was used to capture single–kernel RGB images totaling to 100 samples for each variant. RGB color space thresholding was used in seed vegetation. 22 spectro–textural–morphological features were extracted and 4 were selected using feature importance with extra trees classifier (FI–ETC). KNN, decision tree for classification (DTC), Naïve Bayes (NB), and SVM with color, texture, and morphological seed features as inputs were configured to classify the lettuce seed cultivar. DTC and SVM bested other machine learning models in classifying lettuce seeds with accuracy and sensitivity of 100% using cross and holdout validation. DTC exhibited the fastest inference time with SVM lagging 48.157% behind DTC. This developed hybrid FI–ETC–DTC model is useful for correctly sorting of seeds necessary for controlled–environment cultivation and seed breeding.