E2-YOLO: Enhanced fault detection for smart meter calibration systems

Addressing the challenges of meter malfunction detection caused by uneven ambient lighting and surface reflection interference, this paper presents E2-YOLO, an enhanced fault diagnosis method for single-phase and three-phase meter automated calibration systems. To overcome issues such as non-uniform illumination, small-scale faults, and indistinct fault boundaries, E2-YOLO integrates image data with electrical parameters to improve both fault detection and classification performance. The proposed method introduces several key innovations, including a task-specific multi-branch feature extraction backbone that jointly captures spatial details and temporal dynamics, a multi-scale Retinex-based adaptive illumination correction module designed for industrial camera environments, a bidirectional attention-enhanced feature pyramid network for robust multi-scale fault representation, and a dual-attention mechanism that combines channel and spatial attention to improve small-scale fault localization. Experimental results demonstrate that E2-YOLO significantly improves fault identification accuracy, localization precision, and overall system robustness, providing a reliable solution for industrial meter calibration applications in smart grid environments.