Deep learning-enabled real-time acoustofluidic sorting for selective microalgae recovery under wastewater-relevant heterogeneous conditions.

Microalgae-based wastewater valorization has attracted increasing attention due to its potential for simultaneous pollutant removal and high-value biomass recovery. However, the morphological similarity among algal species and the interference of abiotic particles in heterogeneous wastewater environments severely limit the efficiency of conventional separation approaches. In this study, a real-time intelligent acoustofluidic sorting system integrating image recognition and surface acoustic wave manipulation was developed for selective microalgae recovery under wastewater-relevant heterogeneous conditions. By combining multi-scale feature learning with spatial attention mechanisms, the proposed framework improved the recognition robustness of dynamically flowing microalgae under wastewater-relevant heterogeneous conditions using a controlled microalgal mixture. Under optimized operational conditions-specifically, a sample flow rate of 3 μL/min, a sheath flow rate of 1 μL/min, and an acoustic excitation power of 19 dBm -the system exhibited a perception-to-command latency of 9.8 ms, defined as the interval from image acquisition and target detection to the issuance of the acoustic actuation command, enabling synchronized acoustofluidic sorting under the optimized laboratory conditions. Experimental results demonstrated that the purity of Haematococcus pluvialis in mixed samples increased from 68.5% to 96.6% after sorting, while cellular metabolic activity was effectively preserved. The proposed strategy provides a promising approach for high-purity microalgae recovery and intelligent bioresource recycling in heterogeneous environmental systems.