Digital image analysis coupled with artificial neural networks for simultaneous prediction of biomass and pigments in microalgal cultivation.

Real-time, non-destructive monitoring of multiple physiological parameters in microalgal cultures remains a significant analytical challenge, as conventional methods are destructive, time-consuming, and unsuitable for in situ applications. This study developed a novel digital image analysis framework integrating five color spaces (RGB, HSI, HSV, L*a*b*, YCbCr) with two distinct modeling platforms-Artificial Neural Networks (ANN) and Response Surface Methodology (RSM)-for the simultaneous prediction of biomass (optical density, OD₇₅₀) and key pigments (chlorophyll a, chlorophyll b, and total carotenoids) in Dunaliella salina cultures subjected to combined salinity and light stress. Validation using an independent cultivation dataset demonstrated that the optimal modeling approach was dictated by the physiological nature of the target parameter: ANN models significantly outperformed RSM for non-linear, stress-induced responses, with the ANN-RGB model achieving the best carotenoid prediction (MSE: 0.507, R²: 0.918) and the ANN-L*a*b* model excelling for chlorophyll a (MSE: 0.252, R²: 0.814), whereas a simpler RSM-YCbCr model sufficed for chlorophyll b (MSE: 0.514, R²: 0.670). Novel temporal error analysis (CDF and Heatmap) further revealed the superior stability of ANN models throughout the full cultivation cycle. This low-cost, image-based AI framework offers a robust, non-invasive tool for real-time monitoring in microalgal bioprocessing and smart farming applications.