Aiming at the problems of dogs attacking, an improved YOLOv8n-based model for aggressive dog breed recognition is proposed to enhance recognition accuracy and real-time performance. First, a dataset of aggressive dog breeds containing various complex scenes and lighting conditions is constructed, covering 10 common banned dog breeds, providing high-quality samples for model training. Second, the YOLOv8n network is optimized: the GSConv module is introduced into the Backbone to significantly reduce computation and improve feature extraction capabilities; the VoV-GSCSP module is introduced into the Neck to optimize feature fusion strategies; and the EIoU loss function is adopted to accelerate model convergence and improve bounding box prediction accuracy. Experimental results show that the improved model outperforms the original YOLOv8n model in terms of accuracy, mAP50, and mAP50:95, and demonstrates higher recognition accuracy and comprehensive performance compared with other mainstream object detection models.