Dual-domain feature decoupling: a hybrid network for deblurring and detecting low-voltage meters

Low-voltage meter images captured in real-world scenarios often suffer from motion blur and complex backgrounds, which severely degrade the performance of downstream detection tasks. To address this challenge, we propose a hybrid network named Dual-Domain Feature Decoupling Network (DDFD-Net), which jointly performs image deblurring and meter detection. The core idea is to decouple and leverage complementary information from both the spatial and frequency domains. Specifically, a dual-branch encoder extracts low-level spatial features and high-frequency information in parallel, enabling robust feature representation under blurred conditions. To enhance the model's ability to capture global context and local details, a frequency-aware self-attention mechanism is embedded in the decoder. Furthermore, a task-specific detection head is introduced to predict bounding boxes of key components on the deblurred images, making the framework suitable for real-time meter reading systems. Experimental results on a synthetic and real-world low-voltage meter dataset demonstrate that our approach significantly outperforms existing deblurring and detection baselines in both accuracy and speed, achieving efficient dual-task performance with strong generalization ability.