摘要
为优化智能汽车环境感知算法,提出一种改进的YOLOv8模型。通过引入Ghostconv改进C2f结构,实现算法的轻量化;建立了基于CBAM(Convolutional Block Attention Module)注意力机制的GSAM注意力机制降低模型的计算复杂度,同时提升检测精度;利用P2小目标检测层和深度可分离卷积改进Neck层,降低参数量和计算量提高检测精度;采用SIOU损失函数,提高目标检测算法的性能。实验结果显示,改进后的模型mAP提高了2.5%,复杂度降低了1.4%。
In order to optimize the intelligent vehicle perception algorithm,an improved YOLOv8 model was proposed.The algorithm was lightened by introducing the Ghostconv-enhanced C2f structure.Furthermore,the GSAM attention mechanism was established based on the Convolutional Block Attention Module(CBAM)to reduce the model's computational complexity while enhancing detection accuracy.The model's parameter and computational complexities were reduced,and detection accuracy was improved through the incorporation of a P2 small target detection layer and depthwise separable convolution in the Neck layer.Additionally,the SIOU loss function was employed to enhance the performance of the object detection algorithm.Experimental results demonstrated a 2.5% increase in mAP and a 1.4%reduction in complexity for the improved model.
作者
徐福良
赵红
罗勇
路来伟
XU Fuliang;ZHAO Hong;LUO Yong;LU Laiwei(College of Mechanical and Electrical Engineering,Qingdao University,Qingdao 266071,China)
出处
《青岛大学学报（工程技术版）》
CAS
2024年第1期81-86,94,共7页
Journal of Qingdao University(Engineering & Technology Edition)
基金
青岛市民生科技计划资助项目(19-6-1-88-nsh)。
