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

Area of Science:

• Biotechnology
• Environmental Engineering
• Artificial Intelligence

Background:

• Anaerobic digestion is a key low-carbon technology for biomass utilization and energy recovery.
• Accurate monitoring of fermentation stages is crucial for optimizing anaerobic digestion processes.
• Current methods may lack the precision and automation needed for efficient process control.

Purpose of the Study:

• To develop and validate an advanced computer vision model for detecting anaerobic digestion fermentation stages.
• To enhance the accuracy and automation of visual monitoring in anaerobic digestion.
• To contribute to more efficient and sustainable bioenergy production.

Main Methods:

• Proposed a novel detection method based on the SFN-YOLO model.
• Integrated spatial feature interaction fusion pyramid network (SFIFPN) for enhanced feature extraction.
• Employed a focused linear attention (FLAtt) mechanism to concentrate on critical visual features.
• Utilized normalized Wasserstein distance (NWD) loss for robust bounding box regression.

Main Results:

• The SFN-YOLO model achieved high detection performance on anaerobic digestion images.
• Achieved a precision of 91.9%, recall of 77.6%, and mAP@0.5 of 81.1% on the validation dataset.
• Demonstrated a high rate of correctly detected boxes, reaching 84.27%.

Conclusions:

• The proposed SFN-YOLO method offers a promising approach for intelligent visual monitoring of anaerobic digestion.
• This technology can significantly improve the accuracy, efficiency, and automation of fermentation stage detection.
• The findings support the advancement of sustainable bioenergy management through AI-driven process optimization.