YOLOV12-CMNet: Integrating Cross-domain Attention Fusion for Microorganism Detection in Sewage Water

Le résumé fourni par la source

With the growth of the economy and urbanization, sewage treatment has become increasingly important. Microbial quantity and diversity are key indicators of treatment effectiveness. However, traditional microbial detection methods face challenges like difficulty in observation and high time costs. To address these, deep learning-based image processing for microbial detection offers a promising solution. This paper proposes YOLOv12-CMNet, an improved method based on YOLOv12. We introduce the MAFM module to optimize the backbone network, improving global context awareness without significant computational cost, while reducing sewage background noise. Additionally, the CAFM module is added to model the spatial relationships of microorganisms through a dynamic KNN graph, enhancing feature fusion. Experiments on a custom sewage microbial dataset show that YOLOv12-CMNet achieves 92.3% detection accuracy, with a 3.8% increase in mAP@0.5, a 63.1% speed improvement, and only a 15.1% increase in parameters compared to YOLOv12n. These results demonstrate the effectiveness of YOLOv12-CMNet for intelligent sewage treatment.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.