Bacterial colony counting could be rapid, adaptive and automated

Abstract

Although many attempts have been made to automate bacterial colony counting, little has tackled the counting of clustered colonies and adaptations to handle different bacteria species. In this work, we explore the counting by density estimation method via few-shot learning. We have avoided the difficult localisation and detection of clustered colonies by estimating a density map from the input image. We have also exploited exemplars provided by users to make the method agnostic and adaptive to different bacteria species. Our experiments show that using the counting by density estimation method via few-shot learning results in a promising accuracy from the data set provided by Synoptics Ltd.