Face Detection Study Based on YOLOv11n Improved ViT with AKConv

This study addresses the challenges of high latency, missed detections in high-coverage and multi-target face scenarios, and inefficiencies in traditional face recognition algorithms under complex conditions. An improved YOLOv11 model is proposed, incorporating AKConv with adaptive kernel design to replace standard convolution layers, reducing parameters and computational complexity. Additionally, a ViT module is integrated into the BackBone to enhance global feature extraction via self-attention mechanisms. Experimental results on a custom dataset demonstrate a 97.8% accuracy, with model parameters reduced to 2.3M. Ablation studies confirm the effectiveness of AKConv and ViT in improving detection performance and model efficiency.