Diagnosis of bacterial adhesion performance of activated sludge based on AI-assisted image analysis

ABSTRACT To estimate the bacterial adhesion performance of activated sludge (AS), batch adhesion tests of Escherichia coli (E. coli) on AS and quantitative image analysis (QIA) with a deep learning-based image classifier for magnified AS images were conducted. The pseudo-first-order rate constants for the adhesion and removal of E. coli onto AS resuspended in secondary clarifier effluent, ranging from 0.52 to 1.20 h−1, showed a positive correlation with perimeter-to-area ratio (PAR) (19–47 mm−1) with a Pearson correlation coefficient of 0.82 (p = 0.004). An image classifier determining whether AS is aggregated or dispersed may be used to estimate bacterial adhesion performance, provided that the analyzed image is similar to one of the supervised AS images. However, the rate constant appears to be specifically related to the PAR. In this context, QIA may serve as a more effective diagnostic tool than deep learning-based classification in cases where factors that decisively influence the target diagnostic indices (E. coli removal rate) are not labeled for training. These results highlight the importance of purpose-built training using supervised data appropriately aligned with specific research objectives for the practical implementation of image analysis.