Convolutional Neural Network (CNN) untuk Klasifikasi Penyakit Daun Teh (Camellia Sinensis)

Abstract

Developments in Computer Vision have driven advancements in addressing various problems in economics, healthcare, and agriculture. One of the major contributors is Deep Learning. Deep Learning is a subfield of machine learning that develops algorithms and computational models that focus on learning by complex artificial neural networks with many layers (deep neural networks). One type of algorithm used in Deep Learning is Convolutional Neural Network (CNN). This research explores the ability of CNN to recognize and classify various types of diseases in tea leaves. The architecture used is darknet19 with adam optimizer and the use of 150 epochs, batch size 50, and learning rate 10−4. The dataset used consists of 400 data divided into 4 balance classes, namely Algal Leaf, Anthracnose, Bird Eye Spot, and Healthy. In this study, three types of classification testing were carried out, namely binary classification, multiclass classification with balance data, and multiclass classification with imbalance data. The results of binary classification testing show excellent accuracy, precision, recall, and specificity values, which are 100%. Multiclass classification testing with balance data achieved 95% accuracy with 100% recall for Algal Leaf and Healthy leaves, and 83% recall for Antracnose leaves, and 96% recall for Bird Eye Spot leaves. Multiclass classification testing with imbalance data achieved 85% accuracy with 97% recall for Algal Leaf leaves and 100% for Healthy leaves, 60% recall for Antracnose leaves, and 96% recall for Bird Eye Spot leaves. The evaluation results show that the CNN model is able to classify tea leaf diseases with a high level of accuracy. This gives the potential use of this model in supporting a real-time tea plant disease surveillance system