Classification of Cassava (Manihot sp.) Leaf Variants Using Transfer Learning

Abstract

There are several types of cassava leaves with different characteristics, tastes, and nutritional content. Some people use cassava leaves as a vegetable ingredient for daily consumption as a source of fiber and minerals. However, people often have difficulty identifying the different types of cassava leaves, including cassava leaf variants that are locally referred to as gajah, karet, and mentega. This study aims to use transfer learning to identify the variant of cassava leaves. The Inception v3 architecture was selected to build the classification model. To demonstrate the superiority of transfer learning, the Inception v3 architecture was run with two different weights. The first weight was randomly initialized, while the second weight was taken from pre-trained weights from ImageNet. The experimental results show that the classification accuracy rate using the pre-trained weights reached 95.76%. This indicates that the classification model used in this study is promising and can be used for practical purposes in everyday life.