FlowerMate 2.0: Identifying plants in China with artificial intelligence

Plant identification is essential for human beings. For our ancestors, accurate plant identification meant the ability to locate appropriate food sources, materials for fire, tools, weapons, and medicinal herbs to treat diseases and wounds. This knowledge was crucial for the survival of their families and tribes. Presently, challenges such as global warming and population growth necessitate the selection and development of new energy sources, food crops, and medicinal plants to sustain and enhance our well-being, underscoring the ongoing importance of plant identification. Moreover, accurate identification serves as the foundation of nearly all plant-related disciplines, including phylogeny, biogeography, ecological restoration, resource management, biodiversity conservation, agronomy, forestry, and pharmacy.1 Without accurate plant identification, even the most ambitious research endeavors are effectively baseless.

Traditionally, plant identification has been the purview of taxonomists due to its importance and specialization. However, the capacity of taxonomists alone is insufficient for the demands of plant diversity surveys and the growing identification needs across various disciplines. Identification based on morphological characters requires extensive professional training, experience, and significant financial and material investments. Despite this, the past two decades have seen a global decrease in the funding for taxonomy-related projects and the number of taxonomists to the point where taxonomists humorously consider themselves an endangered species.2

Although artificial intelligence (AI) technology has shown promise in aiding experts in rapid plant identification, the scope and coverage of species identifiable by current models fall short of fulfilling the practical needs of daily life and research. Additionally, while intelligent identification tools developed from image databases exist, their application in scientific research is limited due to the absence of systematic evaluation.3 Furthermore, plant images from these databases, particularly crowd-sourced ones, often exhibit varying quality and identification accuracy.