Plant Classification with Advanced Deep Learning Methods

Identifying plant species often depends on analyzing leaf structures, but this method can be unreliable when leaves are absent due to seasonal factors or affected by diseases. This paper proposes a novel approach of using stem images for plant classification and identification. By incorporating stem-based features, this method provides an alternative when leaves are not available. To further enhance accuracy, This approach combines both stem and leaf images when applicable. The classification framework employs three distinct models: ResNet for initial baseline classification, VGGNet with an SVM classifier for advanced feature extraction, and a Vision Transformer (ViT) model for attention-driven analysis. The predictions from these models are then integrated using ensemble learning techniques to produce more accurate results. The classification performance metrics obtained are as follows: ResNet 98% accuracy, ViT 93 % accuracy.