Detection of Anthracnose on Mango Tree Leaf Using Convolutional Neural Network
Analyn N. Yumang, Christian Joseph N. Samilin, John Christian P. Sinlao
Abstract
Mangoes have been one of the most important products that are being produced mostly within tropical regions here in the Philippines. Anthracnose is the most common and serious disease that can occur on mango crops in the country. It is a disease caused by a fungus called Colletotrichum gloeosporioides, which targets leaves, fruits, twigs, and flowering panicles of the crop. For this study, the researchers' aim is to detect anthracnose disease in mango leaves and classify them as healthy or unhealthy. The system will implement (You Only Look Once, Version 3) YOLOv3, which uses the features learned in Convolutional Neural Network to detect a specific object, live videos, and even lesions of plants. The training package comprises around 80.282% of the photographs, while the trial package contains approximately 19.718% of the images. This is done by randomly splitting the data into two sets. This ratio distribution is frequently used in neural network applications. The PDF format was used on the images with 600 dpi for a better resolution. After training the system it obtained 60.680% mean average precision (mAP), 7.79fps, and a lower total validation loss of 20.93. After training the system and using the confusion matrix an accuracy of 83.33% was obtained.