Classification of crown density and foliage transparency scale for broadleaf tree using VGG-16

Abstract

Crown density and foliage transparency are important parameters for tree crown conditions. Previously, observers carried out crown density and foliage transparency assessments manually, which could be a less efficient process.This research aims to use the VGG-16 deep learning architecture to classify the density and transparency of broadleaves tree crowns. In this study, broadleaves tree crown datasets were collected for four types of broadleaves tree: cacao (theobroma cacao), durian (durio zibethinus), rubber (havea brasiliensis), candlenut (aleurites moluccana); then the data is labeled based on the crown density and foliage transparency scale card. The research applies resize and augmentation preprocessing. The model training process uses a scenario of 80% train data, 10% test data, and 10% validation data. After training using the VGG-16 model, the test results showed impressive accuracy, with the highest accuracy reaching 98.40% for candlenut trees, rubber (96.00%), cacao (92.00%), and durian (86.60%). This research shows quite good results in classifying the scale of crown density and foliage transparency with four types of broadleaves tree (cacao, durian, rubber and candlenut) using VGG-16.