Deep learning algorithms have shown great potential in automotive seal defect detection, but challenges remain, such as model complexity and deployment difficulties. Therefore, FSLW-YOLOv8n, a lightweight algorithm for automotive seal defect detection based on an improved YOLOv8n, is proposed in this paper. First, the Bottleneck structure in the C2f module is optimized by introducing the Faster block, which improves memory access efficiency and feature extraction capabilities. Meanwhile, the neck network adopts the design concepts of GSConv and Slim-neck, significantly reducing the parameter count to achieve model lightweight. Additionally, LADH-Head, a lightweight asymmetric decouple head, is used to further streamline the model structure while improving detection accuracy. Then, the Wise-IoU loss function is introduced, enhancing overall detection performance by a refined small-object localization strategy. Finally, the improved algorithm is converted and deployed on the HiSilicon platform, followed by performance validation. Experimental results show that, compared to the baseline model, mAP has increased by 2.1%, while calculation amount, parameter count, and model size have decreased by 55.6%, 42.7%, and 38.3%, respectively. On the HiSilicon SD3403 embedded platform, the detection speed reaches 31.3 f/s.