Implementation of Convolutional Neural Network for Classification of Density Scale and Transparency of Needle Leaf Types

Abstract

Crown density and transparency are among the parameters in determining forest health using magic card. This is still less effective because it only relies on direct vision. Therefore, a more sophisticated and accurate application using digital image technology is needed. Convolutional Neural Network (CNN) is designed to help recognize objects in images with various positions. There are 1000 images of needle leaf types with ten classes of crown density and transparency for every kind of needle leaf, including araucaria heterophylla, cupressus retusa, pine merkusii, and shorea javanica, which are classified using AlexNet. AlexNet is a CNN architecture that has eight feature extraction layers. The AlexNet model succeeded in classifying coniferous trees on the scale of density and crown transparency with an accuracy level of 87.00% for araucaria heterophylla, cupressus retusa 96.00%, merkusii pine 86.00%, and shorea javanica 95.00%. Although some errors were still found in classification, this was caused by similar patterns and similar image positions. It is hoped that the results of this research will be used in monitoring forest health in the future.