Comparison of CNN Models Using Transfer Learning and an Ensemble Approach for Plant Species Classification

Classifying plant species is an important part of environmental monitoring, biodiversity preservation, and forestry management. However, manually inspecting every plant could be laborious and time-consuming. Convolutional neural networks (CNNs) have shown incredible promise for this problem. The goal of this project is to employ artificial intelligence and machine learning to address this problem. Using transfer learning, we evaluate how well several pretrained CNN models perform in classifying tree species from the pl@nt300k dataset. The classification accuracy, computational efficiency, and flexibility of models like EfficientNetb0, ResNet50, InceptionV3, DenseNet121, and Xception are assessed using a dataset with 24050 photos, with EfficientNetb0 achieving the best accuracy of 90.94% out of all the pretrained model. Furthermore, we provide two ensemble approach using soft voting and weighted average model, achieving an accuracy of 96% and 95% respectively. Results show that the proposed ensemble model provides a more effective classification model for the identification of plant species. We have also discussed the architecture of the hybrid model and how it benefits the classification process, which could be of help in further research to create more effective, flexible and scalable models.