Classification of Succulent Plants Utilizing Vision LLMs

Automated plant classification has been transformed by machine learning techniques, especially using vision-language models. Given their physical diversity and apparent resemblance to other species, succulent plants present a difficult categorization challenge. This paper presents a method using Vision Language Models (Vision LLMs) for the classification of succulent plants based on the "Cactus Aerial Photos" dataset from Kaggle as a benchmark. There are over 16,000 annotated aerial images of several succulent cactus species in this dataset. We extract semantically rich features straight from plant images using a pre-trained Vision Transformer-based LLM with zero-shot and fine-tuned capabilities, therefore removing the need for manually created pipelines or convolutional backbones. The 99.77% classification accuracy of the proposed model shows how well Vision LLMs perform in high-fidelity plant identification applications. Particularly in low-supervision and domain-transfer environments, our work highlights the feasibility of large multimodal models in botanical classification.