基于改进BiSeNet的非结构化道路分割算法研究

宋亮, 谷玉海, 石文天

宋亮, 谷玉海, 石文天. 基于改进BiSeNet的非结构化道路分割算法研究[J]. 应用光学, 2023, 44(3): 556-564. DOI: 10.5768/JAO202344.0302003
引用本文: 宋亮, 谷玉海, 石文天. 基于改进BiSeNet的非结构化道路分割算法研究[J]. 应用光学, 2023, 44(3): 556-564. DOI: 10.5768/JAO202344.0302003
SONG Liang, GU Yuhai, SHI Wentian. Unstructured road segmentation algorithm based on improved BiSeNet[J]. Journal of Applied Optics, 2023, 44(3): 556-564. DOI: 10.5768/JAO202344.0302003
Citation: SONG Liang, GU Yuhai, SHI Wentian. Unstructured road segmentation algorithm based on improved BiSeNet[J]. Journal of Applied Optics, 2023, 44(3): 556-564. DOI: 10.5768/JAO202344.0302003

基于改进BiSeNet的非结构化道路分割算法研究

基金项目: 北京市科技委促进高校内涵发展-学科建设专项资助项目(5112011015);机电系统测控北京市重点实验室开放课题资助(KF20202223204)
详细信息
    作者简介:

    宋亮(1999—),男,硕士研究生,主要从事深度学习和无人车定位与导航研究。E-mail:3028626195@qq.com

    通讯作者:

    谷玉海 (1976—) ,男,博士,研究员,主要从事测控技术及仪器研究。E-mail:gyuhai@163.com

  • 中图分类号: TN99;TP183

Unstructured road segmentation algorithm based on improved BiSeNet

  • 摘要:

    非结构化道路通常没有清晰的边界及车道线,环境较为复杂,传统的基于道路纹理、颜色特征的分割方法无法满足实时性和准确性的要求。针对非结构化道路场景,提出了基于改进BiSeNet的轻量化语义分割模型,采用轻量化主干提取网络和引入深度可分离卷积,优化速度控制;在最后的特征融合阶段引入通道注意力,自适应地选择重要特征,抑制冗余信息,提高非结构化道路分割的准确性。改进后模型参数量仅有1.11×106,检测速度提升18.83%,F1-score达到了96.74%。对比其他主流语义分割模型,该算法具有参数量小、速度快、准确率高等优势,可为非结构化道路场景下无人驾驶车辆的安全运行提供参考。

    Abstract:

    Unstructured roads usually have no clear boundaries and lane lines, and the environment is more complex. The traditional segmentation methods based on road texture and color features cannot meet the requirements of real-time performance and accuracy. For unstructured road scenes, a lightweight semantic segmentation model based on improved BiSeNet was proposed, which adopted the lightweight trunk extraction network and introduced the depthwise separable convolution to optimize the speed control. The channel attention was introduced in the final feature fusion stage to adaptively select important features, suppress redundant information, and improve the accuracy of unstructured road segmentation. The number of parameters of the improved model is only 1.11×106, the detection speed is increased by 18.83%, and the F1-score reaches 96.74%. Compared with other mainstream semantic segmentation models, the proposed algorithm has the advantages of small parameters, high speed and high accuracy, which can provide a reference for the safe operation of unmanned vehicles in unstructured road scenarios.

  • 图  1   BiseNet网络结构

    Figure  1.   Structure diagram of BiseNet network

    图  2   深度可分离卷积

    Figure  2.   Schematic of depthwise separable convolution

    图  3   改进后的空间路径

    Figure  3.   Schematic of improved spatial path

    图  4   Bneck结构图

    Figure  4.   Structure diagram of Bneck

    图  5   改进后的上下文路径

    Figure  5.   Schematic of improved context path

    图  6   通道注意力特征融合模块

    Figure  6.   Schematic of channel attention feature fusion module

    图  7   IDD数据集类别重新划分示意图

    Figure  7.   Schematic of category reclassification of IDD dataset

    图  8   预处理结果图

    Figure  8.   Preprocessing results graph

    图  9   F1-score和loss训练过程曲线变化图

    Figure  9.   Variation curves of F1-score and loss during training

    图  10   预测效果对比图

    Figure  10.   Comparison chart of prediction effect

    表  1   调整后的MobileNetV3-large特征提取网络

    Table  1   MobileNetV3-large feature extraction network after adjustment

    Input(H2×C操作nSE激活函数sout
    5122 × 3Conv2d1-HS216
    2562 × 16Bneck,3 × 31-ReLU116
    2562 × 16Bneck,3 × 31-ReLU224
    1282 × 24Bneck,3 × 31-ReLU124
    1282 × 24Bneck,5 × 51ReLU240
    642 × 40Bneck,5 × 52ReLU140
    642 × 40Bneck,3 × 31-HS280
    322 × 80Bneck,3 × 33-HS180
    322 × 80Bneck,3 × 32HS1112
    322 × 112Bneck,5 × 51HS2160
    162 × 160Bneck,5 × 52HS1160
    表中:HC表示特征的边长、通道数;Conv2d表示标准卷积;n表示需进行操作的次数;SE表示是否引入SE注意力;s表示步长;out表示输出通道数。
    下载: 导出CSV

    表  2   不同改进措施的模型性能

    Table  2   Model performance with different improvement measures

    模型MIOU/%F1-score/%Params
    Conv+Xception+FFM89.8392.661.54×106
    Conv+ResNet+FFM91.9394.321.29×107
    Conv+MobileNetv3+FFM92.4394.651.16×106
    DSConv+MobileNetv3+FFM92.6894.801.09×106
    DSConv+MobileNetv3+CAFFM95.4096.741.11×106
    下载: 导出CSV

    表  3   不同网络模型性能对比

    Table  3   Performance comparison of different network models

    模型F1-Score/%PA/%MIOU/%ParamsSpeed/
    f·s−1
    U-Net91.5695.6789.381.94×10626.17
    DeepLabV3+95.8597.4694.233.87×10715.21
    FCN-32S92.5496.1890.565.44×10728.63
    BiSeNet-resnet94.3296.2791.934.92×10723.57
    BiSeNet-xception92.6695.3189.831.54×10630.96
    本文96.7497.7895.401.11×10636.79
    下载: 导出CSV
  • [1] 周继苗, 李必军, 陈世增. 一种多层特征融合的道路场景实时分割方法[J]. 测绘通报,2020,514(1):10-15.

    ZHOU Jimiao, LI Bijun, CHEN Shizeng. A real-time segmentation method of road scene based on multi-layer feature fusion[J]. Bulletin of Surveying and Mapping,2020,514(1):10-15.

    [2]

    DENG L, YANG M, LIANG Z, et al. Fusing geometrical and visual information via superpoints for the semantic segmentation of 3D road scenes[J]. Tsinghua Science and Technology,2020,25(4):498-507. doi: 10.26599/TST.2019.9010038

    [3]

    DONG G, YAN Y, SHEN C, et al. Real-time high-performance semantic image segmentation of urban street scenes[J]. IEEE Transactions on Intelligent Transportation Systems,2020,22(6):3258-3274.

    [4]

    PAZ D, ZHANG H, LI Q, et al. Probabilistic semantic mapping for urban autonomous driving applications[C]//2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Las Vegas, USA: IEEE, 2020: 2059-2064.

    [5] 黄俊, 侯北平, 董霏, 等. 基于方向纹理的非结构化道路消失点检测研究[J]. 图学学报,2019,40(1):131-136.

    HUANG Jun, HOU Beiping, DONG Fei, et al. Research on vanishing point detection of unstructured road based on directional texture[J]. Chinese Journal of Graphics,2019,40(1):131-136.

    [6] 郭昕刚, 王佳, 程超. 层次聚类算法和基于图的分割算法相融合的图像分割算法[J]. 国防科技大学学报,2022,44(3):194-200. doi: 10.11887/j.cn.202203023

    GUO Xingang, WANG Jia, CHENG Chao. Image segmentation algorithm for the fusion of hierarchical clustering algorithm and graph-based segmentation algorithm[J]. Journal of National University of Defense Technology,2022,44(3):194-200. doi: 10.11887/j.cn.202203023

    [7] 王翔, 张娟, 方志军. 基于最大熵和边缘信息的非结构化道路检测[J]. 电子科技,2020,33(1):23-28.

    WANG Xiang, ZHANG Juan, FANG Zhijun. Unstructured road detection based on maximum entropy and edge information[J]. Electronic Science and Technology,2020,33(1):23-28.

    [8]

    LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(4):640-651. doi: 10.1109/TPAMI.2016.2572683

    [9] 张鑫, 姚庆安, 赵健, 等. 全卷积神经网络图像语义分割方法综述[J]. 计算机工程与应用,2022,58(8):45-57. doi: 10.3778/j.issn.1002-8331.2109-0091

    ZHANG Xin, YAO Qing’an, ZHAO Jian, et al. A review of fully convolutional neural network image semantic segmentation methods[J]. Computer Engineering and Applications,2022,58(8):45-57. doi: 10.3778/j.issn.1002-8331.2109-0091

    [10] 曹富强, 王明泉, 张俊生, 等. 基于深度学习的铸件X射线图像分割研究[J]. 应用光学,2021,42(6):1025-1033. doi: 10.5768/JAO202142.0602003

    CAO Fuqiang, WANG Mingquan, ZHANG Junsheng, et al. X-ray image segmentation of castings based on deep learning[J]. Journal of Applied Optics,2021,42(6):1025-1033. doi: 10.5768/JAO202142.0602003

    [11] 汪水源, 侯志强, 王囡, 等. 基于自适应模板更新与多特征融合的视频目标分割算法[J]. 光电工程,2021,48(10):21-30.

    WANG Shuiyuan, HOU Zhiqiang, WANG Nan, et al. Video target segmentation algorithm based on adaptive template update and multi-feature fusion[J]. Opto-Electronic Engineering,2021,48(10):21-30.

    [12] 王龙飞, 严春满. 道路场景语义分割综述[J]. 激光与光电子学进展,2021,58(12):44-66.

    WANG Longfei, YAN Chunman. Overview of semantic segmentation of road scenes[J]. Advances in Lasers and Optoelectronics,2021,58(12):44-66.

    [13]

    YU C, WANG J, PENG C, et al. Bisenet: Bilateral segmentation network for real-time semantic segmentation[C]//Proceedings of the European conference on computer vision (ECCV). Munich, Germany: Springer-Verlag Berlin, 2018: 325-341.

    [14]

    CHEN L C, PAPANDREOU G, KOKKINOS I, et al. Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,40(4):834-848.

    [15]

    PENG C, ZHANG X, YU G, et al. Large kernel matters—improve semantic segmentation by global convolutional network[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). New York, USA: IEEE, 2017: 1743-1751.

    [16]

    YANG M Y, KUMAAR S, YE L, et al. Real-time semantic segmentation with context aggregation network[J]. ISPRS Journal of Photogrammetry and Remote Sensing,2021,178(12):124-134.

    [17]

    HOEARD A, SANDLER M, CHU G, et al. Searching for mobilenetv3[C]//Proceedings of the IEEE/CVF international conference on computer vision. Los Alamitos, USA: IEEE , 2019: 1314-1324.

    [18]

    VARMA G, SUBRAMANIAN A, NAMBOODIRI A, et al. IDD: a dataset for exploring problems of autonomous navigation in unconstrained environments[C]//2019 IEEE Winter Conference on Applications of Computer Vision (WACV). New York, USA: IEEE, 2019: 1743-1751.

    [19] 郭列, 张团善, 孙威振, 等. 融合空间注意力机制的图像语义描述算法[J]. 激光与光电子学进展,2021,58(12):329-338.

    GUO Lie, ZHANG Tuanshan, SUN Weizhen, et al. Image semantic description algorithm integrating spatial attention mechanism[J]. Advances in Lasers and Optoelectronics,2021,58(12):329-338.

    [20]

    JADON S. A survey of loss functions for semantic segmentation[C]//2020 IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB). New York, USA: IEEE, 2020: 1-7.

    [21] 徐聪, 王丽. 基于改进DeepLabv3+网络的图像语义分割方法[J]. 激光与光电子学进展,2021,58(16):225-232.

    XU Cong, WANG Li. Image semantic segmentation method based on improved DeepLabv3+ network[J]. Advances in Laser and Optoelectronics,2021,58(16):225-232.

  • 期刊类型引用(6)

    1. 陈祥雪,付子亲,王凤超,陈进,杨晶. 类H型结构的太赫兹带阻滤波器. 中国光学(中英文). 2024(04): 757-763 . 百度学术
    2. 薛钊,张海婷,杨茂生,宋效先,张晶晶,叶云霞,任云鹏,任旭东,姚建铨. 基于图形化石墨烯的可调谐宽光谱太赫兹吸收器的研究. 激光与光电子学进展. 2022(05): 12-18 . 百度学术
    3. 刘文,田晋平,杨荣草. 三层石墨烯组成的太赫兹可调宽频带超材料吸收器. 测试技术学报. 2022(05): 384-390 . 百度学术
    4. 高万,王建扬,吴倩楠. 基于双金属环的超材料太赫兹宽频带通滤波器的设计与研究. 激光与光电子学进展. 2021(05): 229-236 . 百度学术
    5. 初启航,杨茂生,陈俊,曾彬,张海婷,宋效先,叶云霞,任云鹏,张雅婷,姚建铨. 可调控的太赫兹多频带吸收器特性. 中国激光. 2019(12): 306-312 . 百度学术
    6. 米洋,吴倩楠,闫仕农. 多频带太赫兹滤波器的设计. 应用光学. 2016(05): 759-764 . 本站查看

    其他类型引用(1)

图(10)  /  表(3)
计量
  • 文章访问数:  325
  • HTML全文浏览量:  94
  • PDF下载量:  69
  • 被引次数: 7
出版历程
  • 收稿日期:  2022-07-04
  • 修回日期:  2022-09-28
  • 网络出版日期:  2023-02-06
  • 刊出日期:  2023-05-14

目录

    /

    返回文章
    返回