Performance analysis of deep learning models for tree species identification from UAV images

This paper addresses the crucial task of identifying tree species, which is often intricate, time-consuming, and reliant on expert knowledge. To tackle this challenge, it employs cutting-edge technologies such as deep learning, particularly focusing on the convolution mixer model, and leverages unmanned aerial vehicle (UAV) images for the identification process. The study concentrates on five distinct tree species such as banana, coconut, mango, oil palm, and papaya. The research uses a dataset of 4200 images, with 80% used for training and the remaining 20% used for testing, and compares the performance of the convolution mixer with other well-known deep learning models like VGG16, MobileNetV2, InceptionV3, and ResNet50. The performance of each model is obtained in terms of mean precision, mean recall, mean F1-score, and overall accuracy. Significantly, the convolution mixer emerges as the standout performer, achieving an impressive overall accuracy rate of 98.22%. This remarkable accuracy highlights the convolution mixer’s potential to revolutionize the field of tree species identification, especially when applied to UAV images. It not only streamlines the process but also ensures high accuracy, promising to aid conservation efforts, biodiversity studies, and sustainable forestry practices.