基于KANs网络和注意力机制的车道线检测方法研究  

Research on Lane Line Detection Method Based on KANs Network and Attention Mechanism

作　　者：杨潞霞 陆剑 YANG Lu-xia;LU Jian(Taiyuan Normal University,Jinzhong 030600,China)

机构地区：[1]太原师范学院,山西晋中030600

出　　处：《电脑与电信》2025年第1期32-36,41,共6页Computer & Telecommunication

基　　金：山西省重点研发计划,项目编号:202102010101008。

摘　　要：车道线检测对于自动驾驶和辅助驾驶系统的安全性至关重要。然而,在光照变化、车道线模糊、车道线部分缺失、阴影等多种复杂环境下,其检测精度会受到较大影响。针对上述问题,以UNet作为主干网络,提出了一种基于KANs网络和注意力机制的车道线检测方法研究。首先在编码网络的卷积层中提出双卷积残差特征提取模块,延缓梯度消失问题;其次设计了HLA注意力机制模块,并将其添加到跳跃连接处,以提升模型的鲁棒性;最后将KAN层集成到UNet模型的深层网络中,提高模型的可解释性。在CULane通用数据集上的测试结果表明,改进后的模型与原模型相比,F1值提高了2.7%,验证了模型改进的有效性。Lane markings detection is critical to the safety of automated driving and driver assistance systems.However,in a variety of complex environments such as lighting changes,blurred lane markings,partial loss of lane markings,and shadows,its detection accuracy will be greatly affected.In order to solve the above problems,this paper takes UNet as the backbone network and proposes a lane line detection method based on KANs network and attention mechanism.Firstly,a double convolutional residual feature ex‐traction module is proposed in the convolutional layer of the coding network to delay the gradient disappearance.Secondly,the HLA attention mechanism module is designed and added to the jump connection to improve the robustness of the model.Finally,the KAN layer is integrated into the deep network of the UNet model to improve the interpretability of the model.The test results on the CU‐Lane general dataset show that the F1 value of the improved model is increased by 2.7%compared with the original model,which verifies the effectiveness of the model improvement.

关 键 词：注意力机制 残差模块 车道线检测 深度神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] U463.6[自动化与计算机技术—计算机科学与技术]