The wireless charging system of EV is easily affected by foreign objects, resulting in a diminished charging efficiency, even triggering a series of safety issues. Consequently, detecting foreign objects in EV wireless charging systems has emerged as a pressing issue requiring urgent attention. This paper investigates a method for detecting foreign objects specifically tailored for EV wireless charging systems. This method employs an improved YOLOv8 model, which is modified from the original YOLOv8 model. Firstly, the backbone network of the YOLOv8 model is replaced with the PP-LCNet model. Secondly, deformable convolutions are introduced into the backbone network layers. Subsequently, an SENet module incorporating an attention mechanism is integrated into the neck network section. Ultimately, the initial loss function is superseded by the Wise-IoU boundary loss function. This paper validates the proposed method through simulation. For the collected foreign object dataset, the enhanced model exhibits a 26% decrease in the number of parameters and a 13% reduction in floating-point operations when compared to the original YOLOv8 model, while achieving a 3.1% increase in accuracy over the same baseline.