Fusion of CNN Models and Optimizers for Plant Species Identification Using Deep Learning

Abstract

Plant species identification plays a crucial role in biological research, ecosystems, and related studies. With recent advancements in artificial intelligence and deep learning, these technologies provide powerful alternatives to manual plant identification. However, previous research lacks comprehensive studies on the combination of CNN models and optimizers for plant species identification. In this study, convolutional neural network (CNN) models, namely ResNet50, VGG16, and EfficientNetB0, were utilized for plant species classification, and their performance was further investigated by integrating different optimizers, including Adam, SGD, and RMSProp. The models were evaluated on two fern species, namely Nephrolepis biserrata and Nephrolepis cordifolia. The dataset comprises 360 images for each class. The experimental results reveal that the fusion of EfficientNetB0 with Adam achieved the highest accuracy of 95.59%. Overall, based on average performance across optimizers, EfficientNetB0 proved to be the most effective model, while RMSProp emerged as the most consistent optimizer. These findings demonstrate that the combination of EfficientNetB0 and Adam is particularly suitable for plant species classification.