Real-time monitoring of activated sludge flocs via enhanced mask region-based Convolutional Neural networks.

ABSTRACT

The functionality of activated sludge in wastewater treatment processes depends largely on the structural and microbial composition of its flocs, which are complex assemblages of microorganisms and their secretions. However, monitoring these flocs in real-time and consistently has been challenging due to the lack of suitable technologies and analytical methods. Here we present a laboratory setup capable of capturing instantaneous microscopic images of activated sludge, along with algorithms to interpret these images. To improve floc identification, an advanced Mask R-CNN-based segmentation that integrates a Dual Attention Network (DANet) with an enhanced Feature Pyramid Network (FPN) was used to enhance feature extraction and segmentation accuracy. Additionally, our novel PointRend module meticulously refines the contours of boundaries, significantly minimising pixel inaccuracies. Impressively, our approach achieved a floc detection accuracy of >95%. This development marks a significant advancement in real-time sludge monitoring, offering essential insights for optimising wastewater treatment operations proactively.