摘要
针对车道线分割任务中的精度和鲁棒性问题,采用一种融合蛇形卷积与卷积注意力的车道线分割方法,以提高车道线识别准确性和稳定性。在模型设计方面,首先引入蛇形卷积模块,将其嵌入到主干网络中,以增强特征提取能力,并更好地捕捉车道线的形状信息;随后,引入卷积注意力模块,用于动态调整特征图的权重,使模型能够更加聚焦于车道线区域,提升车道线分割精度。为了验证该方法的效果,在BDD100K数据集上进行了实验评估。实验结果表明,与主流模型Hybridenet、YOLOP、YOLOM相比,准确率分别提高了0.8%、1.4%、1.3%,IoU分别提高了1.4%、5.1%、2.8%,证明了蛇形卷积和卷积注意力在车道线分割中的有效性和互补性。
To solve the problem of accuracy and robustness in lane segmentation task,a deep learning model is proposed,which integrates snake convolution and convolution attention mechanism to improve the accuracy and stability of lane recognition.In terms of model design,the snake convolutional module is first introduced and embedded in the backbone network to enhance feature extraction capability and better capture the shape information of the lane lines.Then,the convolutional attention module is introduced to dynamically adjust the weight of the feature map,so that the model can focus more on the lane line area and improve the segmentation accuracy of the lane line.To verify the performance of the proposed fusion model,we performed an experimental evaluation on the BDD100k dataset.The experimental results show that compared with the mainstream models Hybridenet,YOLOP and YOLOM,the accuracy rate increases by 0.8%,1.4% and 1.3%,and the IoU increases by 1.4%,5.1% and 2.8% respectively.This result proves the effectiveness and complementarity of snake convolution and convolution attention in lane line segmentation.
作者
黄忠盛
魏敏
HUANG Zhongsheng;WEI Min(School of Computer Science,Chengdu University of Information Technology,Chengdu 610225,China)
出处
《软件导刊》
2025年第8期82-87,共6页
Software Guide
关键词
蛇形卷积
卷积注意力
车道线分割
实例分割
snake convolution
convolution attention
lane line segmentation
instance segmentation
作者简介
通讯作者:黄忠盛(1999-),男,成都信息工程大学计算机学院硕士研究生,研究方向为图像处理;魏敏(1978-),男,博士,成都信息工程大学计算机学院教授,研究方向为图像处理与目标检测技术、3D仿真技术和虚拟现实应用技术。
