Deep Learning-based Method for Multi-Class Classification of Oil Palm Planted Area on Plant Ages Using Ikonos Panchromatic Imagery

Abstract

Oil palm has many advantages, such as biofuels, cosmetics, food ingredients, etc. The amount of oil contained in oil palm fruit is very dependent on the age of the plant, so automatic detection of oil palm plantation area based on plant ages is required to estimate the amount of oil. The use of high-resolution satellite images in oil palm detection has shown promising results for small dimensions, and previous studies have used more than one band of the satellite images data. This will be a burden in terms of cost and processing. Previous studies regarding oil palm area detection usually focused on detecting land cover to distinguish oil palm and non-oil palm areas. This study proposes a method based on deep-learning convolutional neural networks to classify oil palm plantations at a productive age. The images used in this study are the Ikonos satellite image with panchromatic bands only, which have a spatial ratio of 1m. The plantation area is classified into the non-oil palm, oil palm areas with young, mature, and old ages. This study proposes a multi-class classification method for oil palm plantations based on plant ages using convolutional neural networks (CNN). This study performs two fine-tune models on a pre-trained CNN and then classified using SVM and CNN. The performance of CNN architectures such as AlexNet, VGG16, and VGG19 was compared. The highest accuracy is 94.74% when using the CNN classifier and fine-tune model-2 of the VGG19 pre-trained network.