Abstract: Digital Microfluidics (DMF) is an emerging interdisciplinary technology. It focuses on manipulating discrete liquid droplets at the microscale. This technology offers significant advantages, including low reagent consumption, high analytical sensitivity, and high system integration. However, the intelligent development of current DMF systems faces a major bottleneck. It is highly challenging to recognize multi-state droplets with high accuracy in real time. Because of this limitation, achieving efficient closed-loop feedback control remains a difficult task.To solve this problem, this paper proposes a real-time semantic segmentation model named FANeXt. This model is specifically designed to recognize droplets in DMF devices quickly and accurately. The overall architecture relies on an improved convolutional attention mechanism and a novel foreground perception branch.The proposed model features two main technical contributions. First, we improve the multi-scale convolutional attention module. We introduce learnable weight parameters and local functional branches into the network. These additions help the model dynamically adjust the importance of features extracted from different scales. Second, we design a lightweight Foreground-Aware Branch. This component explicitly models the relationship between the droplet foreground and the chip background. By doing so, it effectively suppresses complex background noise and sharpens the droplet boundaries.Furthermore, this study constructs a comprehensive DMF droplet dataset. This dataset includes various typical motion states of droplets, such as moving, splitting, and merging.Experimental results demonstrate that the FANeXt model performs excellently on the self-built dataset. It successfully achieves an optimal balance between segmentation accuracy (evaluated by mean Intersection over Union, mIoU) and processing speed (evaluated by Frames Per Second, FPS). Ultimately, this research provides a highly efficient visual perception solution for the automation and intelligent advancement of digital microfluidics technology.