Real-time detection and motion analysis of bubbles in gas–liquid two-phase flow via YOLO v11 and DeepSORT

Accurate monitoring of bubble behavior in gas–liquid two-phase flows is vital for the safety and efficiency of processes such as chemical production and nuclear energy. To address the limitations of traditional intrusive methods—low spatiotemporal resolution and reliance on manual tuning—this study proposes a real-time detection and tracking framework integrating You Only Look Once version 11 (YOLOv11) with convolutional block attention module (CBAM) and deep simple online and real-time tracking (DeepSORT). CBAM enhances small-bubble features, and a dynamic anchor box strategy improves localization. The model achieves 98.7% mAP@50 (+1.1%) and 80.6% mAP@50–95 (+2.5%) at 640 × 640 input, with 95% accuracy, 4.2 ms/frame (238 frames per second, FPS), and 16.7% fewer false detections vs baseline YOLOv11n.Experiments using an FASTACAM NOVA S12 high-speed camera (2–6 mm bubbles) and an NVIDIA GeForce Ray Tracing Texel eXtreme (RTX) 4060 Ti GPU confirmed real-time performance. Morphological features—void fraction (0.14–0.44), aspect ratio (1.36–1.61), and interfacial area concentration (0.48–0.85 mm−1)—were analyzed. Continuous trajectory tracking enables precise motion characterization, supporting process control in two-phase flow systems.