In this work, we describe an embedded vision-based system that uses a Raspberry Pi and a CMOS camera to detect and classify bacteria and fungi in Petri dishes in real-time. In order to decrease human interaction and increase diagnosis reliability, this project aims to automate the identification process in microbiological laboratories. Our method makes use of an attention-based filtering mechanism in conjunction with an enhanced YOLOv8 object identification model to improve robustness in the face of demanding laboratory settings, including complicated backdrops and fluctuating lighting. In contrast to current approaches, the suggested system incorporates a real-time Human–Machine Interface (HMI) that facilitates deployment in resource-constrained contexts and offers instantaneous visual feedback. Our system's ability to achieve high detection precision through experimental validation validates the possibility of incorporating deep learning techniques into embedded platforms to improve biological diagnostics’ accuracy and efficiency.