IMPLEMENTASI TRANSFER LEARNING DALAM MENDETEKSI PENYAKIT PADA DAUN GANDUM

Abstract

Wheat is one of the most frequently consumed commodities of Indonesian people. This plant is often consumed as an carbohydrate addition or rice substitution. Most Indonesians process the wheat for ingredients such as flour, bread, instant noodles, cereals and other processed ingredients. Unfortunately, the demand for wheat is not suitable with level of production. One of the factors that hinder wheat production is crop failure due to disease or pests. Diseases that are often found in wheat are Septoria and Stripe Rust. The disease can be identified by color and leaf spot, but it is difficult to distinguish between the two diseases. With the rapid development of technology, this problem can be solved using one of the deep learning techniques known as transfer learning. The purpose of this study was to test five pretrained models to diagnose disease in wheat leaf, the models tested were InceptionV3, MobileNetV2, VGG16, ResNet101V2, DenseNet201. The results of testing and comparing five pretrained models, InceptionV3 gives better results than other models with a low computation time of only 976 seconds or the equivalent of 16 minutes and has a very high accuracy.