Termite Image Classification Using Zero-Shot Learning with Multimodal LLM, FLAVA

Abstract

Jay Kim and Daniel Kim

The classification of termite species is essential for ecological studies, pest management, and biodiversity conservation. However, traditional classification methods require extensive labeled datasets, which are difficult to collect for rare or understudied termite species. This paper presents a novel approach to termite image classification using zero-shot learning (ZSL) with FLAVA, a multimodal foundational model. By leveraging FLAVA’s cross-modal alignment of visual and textual data, we demonstrate its potential to classify termite species without requiring domain-specific fine-tuning. Experimental results on a termite dataset highlight the efficiency and scalability of this approach, setting the stage for broader applications in entomology and ecology.

PDF