Objective In the scenario where pedestrians and electric vehicle drivers share the road an improved electric vehicle helmet-wearing detection algorithm based on Yolov8n is proposed to address the helmet detection of electric vehicle riders. Methods First to improve the model?? s accuracy in detecting low-resolution images and small targets SPDConv was introduced to replace the ordinary convolution and down-sampling operators in the model. Second to enhance the model?? s ability to distinguish objects when the background was similar to the detection target the CG block module was introduced and integrated with the original C2f module to create the C2f-CG module which replaced the original C2f module to boost the context feature extraction capability. Finally to reduce the computational volume of the model and keep the model lightweight the ordinary convolution in the Head layer of the original model was replaced with group convolution. Results After experimental testing the enhanced model proposed in this paper demonstrated an improvement in mAP accuracy of 4. 2% and a reduction in computation of 15% in comparison to the Yolov8n original model. The detection accuracy for low-resolution images and small targets had also increased. Conclusion The improved model is suitable for electric vehicle helmet detection in complex situations and can be used as a solution in practical fields