Frequency-Aware Networks to Robust Low-Voltage Meter Detection in High-light Environments

Low-voltage meter detection under high-light conditions is hindered by reflections, overexposure, and dense layouts, which blur boundaries and degrade accuracy. To tackle this, we propose a frequency-aware feature fusion framework. It includes: (1) a High-Frequency Recovery (HFR) module to restore lost edge and texture details; (2) an Intra-Class Smoothing (ICS) module to reduce intra-meter variation via low-frequency smoothing; and (3) a Dynamic Adaptive Resampling (DAR) module for precise feature alignment in dense regions. Experiments on a custom high-light meter dataset confirm that our method improves detection accuracy and preserves detail under challenging conditions.