2.2.2 - A Lens-free Microscope with Edge AI Acceleration

Event
EUROSENSORS 2026
2026-09-06 - 2026-09-09
Zurich
Band
Lectures
Chapter
Optical Sensors
Author(s)
L. Zhukova, I. Benito-Altamirano - Universitat Oberta de Catalunya,Barcelona (Spain), L. Yan, A. Dharmawan, J. D. Prades - Technische Universität Braunschweig,Braunschweig (Germany)
Pages
83 - 84
DOI
10.5162/eurosensors2026/2.2.2
ISBN
978-3-910600-12-6
Price
free

Abstract

Lens-free microscopy offers a compact lab-on-a-chip solution but typically requires high external computational power for image reconstruction. This work introduces a lens-free system utilizing the Sony IMX500 sensor within the Raspberry Pi AI Camera. By removing the integrated lens, we leverage the sensor's internal AI accelerator to process in-line holographic images directly. Instead of traditional diffraction-based back-propagation, we employ Convolutional Neural Networks (CNNs) for real-time cell counting. Using a synthetic dataset for training, we applied post-training quantization to deploy our model onto the IMX500 hardware. Results demonstrate that the on-sensor AI performs comparably to a GPU, establishing a localized, low-power processing pipeline for portable lab-on-a-chip sensing.

Download