Implementasi Metode Convolutional Neural Network (CNN) untuk Mengklasifikasikan Penyakit pada Kelapa Sawit

Abstract

Oil palm is a production crop that can produce palm oil, which is one of the sources of currency to support oil palm farmers and their families. Therefore, if the oil palm is affected by disease and cannot grow in prime condition, the income of the oil palm farmer can be threatened and reduced. Diseases that often attack oil palms during their growth period include Anthracnose, Yellow Line, and leaf spots. These diseases are troubling because they damage the leaves which are a way for plants to photosynthesize and live. One way to prevent oil palms from dying or growing with disease is to know what diseases infect oil palms at that time so that quick treatment can be done. Therefore, a system is needed that can classify those that are infected by oil palm diseases. The research conducted uses data from oil palm plantation areas in Mukomuko Regency, Bengkulu province. The data used is taken using a cellphone camera, the data taken is 300 palm data and 100 kaggle data, the data is then transposed, flipped, and roteate so that the data becomes 4 times more, namely 1600 total data used. The method used to classify oil palm diseases is Convolutional Neural Networks (CNN) which functions to extract features from oil palm leaf images, and You Only Look Once (YOLO) which is used to detect oil palm objects in images. The data is divided into 4 classes to be entered into the CNN algorithm with each class using 400 amounts of data, and the data is divided into 1120 or 70% training data and 480 or 30% validation data with a total of 1600 datasets. Data is selected and given a bounding box to be entered into the YOLO algorithm which is also divided by the same division. The results obtained from model evaluation are CNN with accuracy 0.9534, Loss 0.0980, Val_accuracy 0.9782, and Val_loss 0.1085, and YOLO with mAP50 0.992, and mAP50-95 0.846.