Feature Sequence Fusion Faster R-CNN-Based Metal Foreign Object Detection for Electric Vehicle Wireless Power Transfer System

The potential safety hazard of metal foreign objects triggering wireless power transfer systems for electric vehicles is gradually attracting attention. Due to the complex charging environment and small foreign object targets, metal foreign objects are difficult to detect. To address the above problems, this study proposes a deep learning metal foreign object detection method for wireless power transfer systems for electric vehicles. First, an image dataset with different backgrounds is made, as well as the data augmentation is applied to process the data to cope with the problem of poor training due to the small number of the dataset. Then, a Feature Sequence Fusion Faster R-CNN target detection framework is proposed to improve the feature extraction ability of the network for small targets. Moreover, the sensitivity for small targets is promoted by improving the Region Proposal Network. Finally, the method is validated in different real parking space scenarios, showing that the method performs well in the task of detecting metal foreign objects with Precision and Recall of 89.3% and 78.9%, respectively.