An SFN-YOLO-based detection method for the fermentation stage in the anaerobic digestion process

Anaerobic digestion is an effective and low-carbon approach for biomass utilization and energy recovery. To accurately determine fermentation stages and enhance process control, this paper proposes an anaerobic digestion fermentation stage detection method based on SFN-YOLO. This model combines spatial feature interaction fusion pyramid network (SFIFPN), a focused linear attention (FLAtt) mechanism that concentrates on key features, and a normalized Wasserstein distance (NWD) loss for robust bounding box regression to accurately detect and classify anaerobic digestion images at different stages. The model demonstrates good detection ability on the validation dataset (precision reaches 91.9%, recall reaches 77.6%, and mAP@0.5 reaches 81.1%), with the total proportion of correctly detected boxes reaching as high as 84.27%. This work demonstrates the potential of intelligent visual monitoring to improve the accuracy, efficiency, and automation of anaerobic digestion processes, contributing to sustainable bioenergy management.