Vision–Detection-Coil Fusion Foreign-Object Detection System for Electric Vehicle Wireless Charging

Foreign-object detection in electric vehicle wireless charging must combine visual anomaly detection with metallic-object identification while limiting the cost of scanning a coil array. This paper proposes a vision–detection-coil fusion framework that selects sensing channels from visual anomalies and evaluates metallic evidence from compensated electrical residuals. A Noisy-OR rule combines the evidence, while ambiguous metallic subtypes and incomplete inspection coverage inhibit charging. For the specified nine-channel scenario, the calculated sweep-and-decision budget is 122.12 ms, compared with 819.52 ms for 64 channels, a reduction of 85.1%. In a simulation of 3600 feature cases with the representative channel observed in every case, four-class accuracy is 82.19%, nonmetallic-class recall is 43.5%, and the object-free false-positive rate is 9.5%. Restricting acquisition to visual candidates gives 48.79% presence recall, equal to vision alone. Independent full-array inspections are therefore required; the illustrative 2-s schedule reserves 40.976% of its time for these scans. Paired comparisons show that dynamic compensation reduces disturbance-induced false alarms under the prescribed additive model. These results characterize the tradeoff between selective acquisition, material classification, and inspection coverage. These numerical evaluations do not establish physical detection limits or thermal safety.