Identification of Plant Species Using Convolutional Neural Network with Transfer Learning

Abstract

Plants are the building blocks of nature and human beings. However, the excessive explosion of population and climate changes, some plants are extinct, and some are on the corner of extinction. Additionally, numerous species remain unexplored till now. Exploring the species in the traditional way are labor-intensive, time-consuming and require specialised expertise. So, it is a very challenging task. To overcome these challenges, various state-of-the-art approaches have been proposed. These approaches often face significant limitations related to accuracy, training and testing processes. This paper proposed a novel approach to species identification leveraging deep learning techniques, employing a weighted average methodology. The proposed approach utilises well known publicly available datasets like Malayakew (MK) and Leafsnap, to evaluate F1 score, recall, accuracy, and precision. In proposed approach we utilised pretrained Convolutional Neural Networks (CNNs) and Transfer Learning (TL) to enhance performance. Specifically, architectures such as NASNet, DenseNet121, ResNet50V2, Xception, VGG19 and VGG16 were employed in the experimental study. The proposed approach achieved an F1 score of 99.9%, recall of 100%, accuracy of 100% and precision of 100% on the MK dataset. On the Leafsnap dataset, the suggested approach achieved an F1 score of 94%, recall of 94%, accuracy of 93.5% and precision of 94%. These results demonstrate that the proposed approach significantly outperforms existing state-of-the-art works, offering a robust and efficient solution for species identification across diverse datasets.