Image Databases for Pollen Grain Classification: Availability, Quality, and Challenges

ABSTRACT Pollen grain identification and categorization have applications in various fields, such as biology, medicine, and food safety. Traditionally, researchers manually perform pollen identification, a time‐consuming process prone to errors. With advancements in artificial intelligence and computer vision, there is growing interest in automating this task to improve efficiency and accuracy, making the availability of high‐quality labeled image databases essential. This study conducted a systematic literature review across six databases (ACM DL, DBLP, ScienceDirect, IEEE Xplore, Springer, and SOL) to map pollen image databases used over the past 10 years. From 1475 initially retrieved studies, rigorous inclusion and exclusion criteria reduced the sample to 73 relevant publications for final analysis. The study analyzed quantitative aspects, such as the number of species and image volume, and qualitative aspects, such as labeling methods and image processing techniques. The analysis identified 18 datasets and three crowdsourced databases, highlighting challenges such as the scarcity of available datasets, geographical concentration of samples, and the need for standardization in image annotation. The results emphasize the importance of expanding dataset diversity and improving image quality to enhance the application of machine learning models in pollen grain classification.