Identification of Tropical Plants Leaves Image Base on Principal Component Analysis

Abstract

Difference and variation of leaves shape is usually used as primary identifier of the plant species. But some plants may have a similar leaf shape and thus require another more accurate identifier. This study applied principal component analysis (PCA) methods for identifying tropical plant species from the shape of the leaves. This method simplified the observed variables by reducing the dimensions of the information that is stored as much as 75%, so it did not eliminate important information and can save the data processing time. There were 100 images of leaves taken from several sides of the leaf in JPEG format with which the shape of leaves were look similar, like citrus (Citrus aurantifolia), durian (Durio zibethinus), guava (Psidium guajava), mango (Mangifera indica), jackfruit (Artocarpus heterophyllus), avocado (Persea americana), rambutan (Nephelium lappaceum), sapodilla (Manilkara zapota), red betel (Piper crocatum) and soursop (Annona muricata). Identification of those 10 kind plant leaves produced 97% accuracy rate. Measurement systems were designed using the K-fold Cross Validation with k = 10, the results of experiments shown omission error occurs on the leaves of guava, jackfruit and red betel while twice commission error were found on the leaves sapodilla and once on citrus leaves.