<p>Detecting the flight trajectory of shuttlecocks in badminton videos faces the dual challenges of complex background interference and real-time processing requirements. To address these issues, we propose a lightweight detection network based on temporal differencing. Specifically, ShuffleNet V2 is employed as a feature encoder to significantly reduce the model’s computational complexity. A novel motion enhancement module is designed to explicitly extract motion cues by adaptively calculating pixel-level differences across consecutive frames. Furthermore, a cascaded channel-deformable attention mechanism is introduced to effectively fuse spatio-temporal features and enhance the algorithm’s detection performance. Experimental results demonstrate that, compared with the baseline network, the proposed method achieves an optimal detection accuracy of 88.5% and an F1 score of 93.3%, while reducing the number of model parameters by 36.8%, thereby validating its effective balance between lightweight design and high detection efficiency.</p>