Transfer learning for quantitative image analysis of biosolids Open Access

This study examines such potential in dewatering of stabilized biosolids from an industrial wastewater treatment plant. The decanters are used for dewatering, with polymers improving separation. The study aims to develop a transparent and systematic analysis workflow encompassing data integration from various sources to predict organic solids recovery. During two campaigns, data were collected from operational conditions, laboratory measurements, and image analysis. Partial least squares (PLS) and random forest (RF) models were tested using different combination of data sources. The campaign results revealed variable correlation between polymer dosage and organic solids recovery due to complex dynamics of solids characteristics originated from biotech production upstream of the decanters. Following clustering of segmented images of individual particles and predicting recovery with a RF model, the presence of specific crystalline particles was found to be significant, linking important recovery dependency to these particles. The best recovery prediction was obtained using a RF model utilizing both process and laboratory data in combination with transfer learning, improving the prediction by 14% as compared to baseline prediction. In general, the RF model outperformed the PLS model in predicting recovery, although both models lack consistency in prediction across the organic solids concentration range.

• A systematic analysis workflow is presented for quantitative image analysis of sludge.

• Partial least squares and random forest models were used to predict solids recovery.

• Total dissolved solids correlated with specific crystalline particles by clustering.

• Transfer learning is a powerful technique to enhance quantitative image analysis.