Implementation of Transfer Learning in the Convolutional Neural Network Algorithm for Identification of Potato Leaf Disease

Abstract

Potatoes are a plant that has many benefits for human life. The potato plant has a problem, namely a disease that attacks the leaves. Disease on potato leaves that is often encountered is early blight and late blight. Image processing is a method that can be used to assist farmers in identifying potato leaf disease by utilizing leaf images. Image processing method development has been done a lot, one of which is by using the Convolutional Neural Network (CNN) algorithm. The CNN method is a good image classification algorithm because its layer architecture can extract leaf image features in depth, however, determining a good CNN architectural model requires a lot of data. CNN architecture will become overfitting if it uses less data, where the classification model has high accuracy on training data but the accuracy becomes poor on test data or new data. This research utilizes the Transfer Learning method to avoid an overfit model when the data used is not ideal or too little. Transfer Learning is a method that uses the CNN architecture that has been trained by other data previously which is then used for image classification on the new data. The purpose of this research was to use the Transfer Learning method on CNN architecture to classify potato leaf images in identifying potato leaf disease. This research compares the Transfer Learning method used to find the best method. The results of the experiments in this research indicate that the Transfer Learning VGG-16 method has the best classification performance results, this method produces the highest accuracy value of 95%.

Klasifikasi Kondisi Ban Kendaraan Menggunakan Arsitektur Vgg16

“…Proses Klasifikasi mendapatkan akurasi tertinggi sebesar 95%. Penelitian ini meyimpulkan bahwa arsitetur VGG16 mendapatkan hasil yang lebih baik (Rozaqi et al, 2021) .…”

Optimization of Potato Leaf Disease Identification With Transfer Learning Approach Using Mobilenetv1 Architecture

“…By performing data augmentation, an accuracy rate of 97% was achieved. A study by (Rozaqi et al, 2021a) classified potato leaves with a dataset of 450 images obtained from PlantVillage (Kaggle). By testing several methods, accuracy results were obtained as follows: InceptionV3 78%, VGG16 95%, and ResNet-50 at 78%.…”