A comparative study on plant classification using convolutional neural networks architectures

Abstract

Determining the species of a plant is important to know its ecological and economic importance. Recently, deep learning (DL) models, specifically convolutional neural networks (CNN), have achieved outstanding results in several applications, including the classification of plants. This work focused on the evaluation and compassion of transfer learning models: Alexnet, VGG-16, ResNet-18, ResNet-50, DenseNet, and Inception V3. The datasets used were the Peruvian Forestry Amazon dataset and PlantVillage. For the training, therefore, we used two instances. We evaluated the models by different multiclass metrics: accuracy, sensitivity, precision, F-score. The results present significant values obtained by the VGG-16 model, with 97,79% accuracy, 98,00% sensitivity, 98,00% precision, and 98,00% F-score to the Peruvian Forestry Amazon dataset. It is possible to conclude that the VGG-16 model got an acceptable level of accuracy, which makes it a useful tool to help classify plant species from the Amazon.