Towards Neuromorphic Event-Based Sensing for High-Speed Multi-Spectral Classification and Tracking of Microparticles
Joana Teixeira, Tomás Lopes, Tiago D. Ferreira, Catarina S. Monteiro, Pedro A. S. Jorge, Nuno A. Silva · 2026
Read Towards Neuromorphic Event-Based Sensing for High-Speed Multi-Spectral Classification and Tracking of Microparticles by Joana Teixeira, Tomás Lopes, Tiago D. Ferreira, Catarina S. Monteiro, Pedro A. S. Jorge, Nuno A. Silva (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-31038.txt.
Introduction
Teixeira and colleagues build a brain-like microfluidic platform for multi-spectral classification and tracking of microparticles. Standard systems trade throughput, speed, and bandwidth, limiting fast-flow tracking without motion blur or huge output. Their platform extracts motion and spectral information from samples by pairing an RGB filter mask with an event-based sensor, coding signature and motion at sensing. It reaches sub-millisecond timing and robust sorting across sizes and speeds, including non-laminar flows, while cutting data bandwidth by over two hundred times versus standard imaging.
Catalog record
| CREATOR | Joana Teixeira, Tomás Lopes, Tiago D. Ferreira, Catarina S. Monteiro, Pedro A. S. Jorge, Nuno A. Silva |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54419 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.31038
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Towards Neuromorphic Event-Based Sensing for High-Speed Multi-Spectral Classification and Tracking of Microparticles Joana Teixeira, Tomás Lopes, Tiago D. Ferreira, Catarina S. Monteiro, Pedro A. S. Jorge, Nuno A. Silva Abstract Conventional image-based microfluidic systems face an inherent trade-off between throughput, imaging speed, and data bandwidth, limiting their ability to monitor high-velocity flows without significant motion blur or prohibitive data generation. Event-based sensing has emerged as a high-speed, low-power alternative, but has so far been largely restricted to tracking monodisperse, spherical particles. In this work, we introduce a microfluidic sensing platform that enables the simultaneous extraction of kinematic and spectral information from polydisperse microparticles using a neuromorphic imaging approach. By integrating a spatially multiplexed RGB filter mask with an asynchronous event-based sensor, spectral signature and motion are encoded directly at the sensing stage, eliminating the need for image reconstruction or learning-based inference. The system achieves sub-millisecond temporal resolution and maintains robust classification performance across a broad range of particle sizes and flow velocities, including under non-laminar conditions, reaching up to 82% accuracy for classification of colored particles within the 0.08–0.18 mm range. The event-driven architecture reduces data bandwidth by > > 240× compared to conventional high-speed imaging, while sustaining an area throughput of 460 mm 2 /s. By providing a computationally efficient and low-latency particle characterization, this framework paves the way for a scalable solution towards high-speed, label-free screening of heterogeneous analytes in clinical diagnostics and environmental monitoring. keywords Neuromorphic Vision, Multi-Spectral Imaging, Microfluidics 1 Introduction In the last decades, microfluidics have become an indispensable tool in a wide range of scientific and technological applications, ranging from drug discovery to single-cell analysis, to name a few examples 13 ; 21 . A critical aspect of these microscale systems is the ability to accurately track and characterize the behavior of particles or cells that flow through microchannels. Currently, image-based microfluidic systems face inherent trade-offs between imaging speed, throughput, and data bandwidth. High-speed CMOS or CCD imaging is often required to capture fast dynamics, but may still lead to motion blur, reduced signal quality, and rapidly increasing data volumes that hinder real-time processing 24 . As a result, increasing the throughput of the system, sometimes crucial for screening large populations, often requires sacrificing either imaging speed, sensitivity, or resolution. Several techniques have been proposed to overcome these limitations, in particular focusing on increasing the throughput of the systems and improving the resulting image quality. For instance, stroboscopic illumination has been successfully integrated into sheathless systems to provide blur-free analysis at throughputs exceeding 50,000 cells/s 15 . By utilizing short, high-intensity light pulses, these systems reduce effective exposure times and increase the depth of field, though they require precise temporal synchronization and high-power light sources. Parallelization via multi-channel architectures or the integration of microlens arrays 7 has further pushed monitoring capacities, offering magnification and enumeration performance comparable to high-end microscopy across a wide field of view, at the cost of an increase in system complexity. More recently, the use of deep learning architectures has demonstrated the ability to enhance or reconstruct low-resolution images 9 ; 16 , but introduces significant computational overhead and may compromise the fidelity of spectral and morphological information. Despite these advances, existing approaches remain fundamentally tied to frame-based imaging and its associated constraints. These technical constraints have motivated the exploration of alternative imaging approaches, such as the use of event-based cameras 8 ; 20 ; 23 . In contrast to adaptations of frame-based architectures, event-based sensing (EVS), also known as neuromorphic vision sensors, may offer a fundamental shift in the imaging paradigm by discarding the concept of global frames altogether. By recording intensity changes asynchronously at the pixel level, EVS can be a powerful alternative for high-speed microfluidic analysis, benefiting from higher dynamic range, microsecond-scale temporal resolution, and promising lower memory and computing overhead 11 ; 2 .In fact, this technology has been demonstrated as a reliable approach for tracking and reconstruction of high-speed fluid flows with velocities up to 1.7\text{\,}\mathrm{m}\text{\,}{\mathrm{s}}^{-1} 20 ; 19 ; 4 . Additionally, the use of event-based cameras has been reported for capturing microparticles moving through microfluidic channels and classifying them according to their size, utilizing advanced neural network architectures 18 ; 23 . By processing the asynchronous event data directly, these deep learning approaches can extract morphological features and categorize the particle size with minimal latency. These results highlight the potential of event streams for high-dimensional characterization. Despite these advances, the exploitation of spectral information from opaque analytes in event-based microfluidic architectures remains largely underexplored. In recent years, and in the context of microscopy, spectral and hyperspectral imaging have attracted increasing interest because wavelength-dependent reflectance and absorption can provide material- and chemistry-specific signatures that are not accessible from morphology alone 22 ; 1 ; 17 . Event-based vision has also begun to be explored for spectral imaging recently, with active illumination schemes demonstrating that spectral information can be encoded into asynchronous event streams while reducing bandwidth requirements 3 . However, to the best of our knowledge, the integration of event-based spectral sensing with continuously flowing samples in a microfluidic channel has not yet been demonstrated. Besides, previous event-based microfluidic studies have largely relied on monodisperse or well-defined particles, such as spherical beads with known geometries, or on learning-based classification pipelines. Such idealized conditions do not fully capture the morphological and optical complexity of real-world samples, where analytes often exhibit substantial heterogeneity in size, shape, and reflectance 14 ; 10 . In this work, we aim to move beyond these constraints by developing a neuromorphic microfluidic sensing platform capable of simultaneous tracking, sizing, and performing color classification of polydisperse fragments. By combining a spatially multiplexed RGB filter mask with an asynchronous event-based sensor, spectral identity and motion are encoded directly at the sensing stage, removing the need for training. We demonstrate real-time tracking, sizing, and color classification under non-laminar flow conditions, establishing a data-efficient and low-latency approach for high-throughput analysis of heterogeneous analytes, and paving the path for future multispectral analytical on-chip solutions. 2 Methods This work combines a microfluidic flow platform, a spatially multiplexed optical filtering strategy, and an event-based processing pipeline to enable simultaneous spectral and kinematic characterization of microparticles. The following sections describe the experimental architecture of the system, including the microfluidic circuit, illumination geometry, and dual-camera imaging configuration, before detailing the calibration and signal-processing procedures used for color prediction, velocity estimation, and size extraction. 2.1 System Architecture and Microfluidics The fluidic circuit is driven by a programmable syringe pump (NE-4002X Programmable 2-Channel, SyringeTWO), which controls the volumetric flow rate of the particle suspension. The suspension is delivered from a 3 ml syringe through flexible tubing into a commercially available microfluidic channel (ibidi \mu -Slide, 0.2 Luer, glass bottom, channel dimensions: 50×5×0.2 mm). The use of a glass-bottom slide ensures high optical transparency and compatibility with high-resolution microscopy. The channel is positioned in an inverted setup, with a 10x objective, and illumination in reflection is provided by a two-white-LED arrangement(see Figure 1 A), enabling effective detection of opaque particles. By collecting backscattered light, the event-based sensor responds to the particles’ intrinsic color and surface reflectance, rather than solely to intensity occlusion. This configuration avoids the ”shadow effect”, where an event camera might