Palm Trees Counting Using MobileNet Convolutional Neural Network in Very High-Resolution Satellite Images

Abstract

Indonesia has a large area of oil palm plantation. Information related to the spatial distribution and number of palm trees is essential for oil palm plantation management and monitoring. The common standard of monitoring the number of oil palm trees has been either manually counting at the plantation itself or from the given aerial images. Manual counting requires many workers and has potential problems related to accuracy. This article presents an approach to the extraction and counting of oil palm trees using deep learning approach. We investigate the use of MobileNet-v1 to detect the individual palm trees from very high-resolution satellite images. MobileNet-v1 is a lightweight CNN architecture model that is usually used on smartphones or other devices with limited processing resources. The network was trained with the dataset that contains 3500 small images of size <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$25\times 25$</tex> pixels. The result shows that this method managed to detect oil palm trees with the precision, recall and F1 score more than 0.9.