Machine Learning As Seed Image Identification Using Principal Component Analysis (PCA)

Abstract

Plants can be identified using several variables, such as seeds’ shapes, colors, and sizes. However, several types of plants have close similarities to seed shapes. Therefore, additional characteristics are necessary to support the identification process. This study applied machine learning with the PCA method to identify plant species from seed shapes. The PCA simplifies the observed variables by reducing data dimensions and storing 75% of the information. The procedure did not eliminate too much important information while reducing data size and processing time. We collected 100 images of plant seeds similar to one another, such as sapodilla seeds, soursop, cucumber, star fruit, grape, melon, apple, lime, watermelon, and chili. A measurement system was designed using the K-Fold Cross Validation, and 10 tables of experimental results discovered a good level of accuracy of 83%. The Omission error occurred in the seeds of soursop, starfruit, grape, apple, lime, and watermelon while the most commission errors occurred in apple seeds (8 times).