1 College of Electric Power, Inner Mongolia University of Technology, Hohhot 010080, China
2 Key Laboratory of Mechanical and Electrical Control at Universities of Inner Mongolia Autonomous Region, Hohhot 010080, China
3 Engineering Research Center of Large Energy Storage Technology, Ministry of Education, Hohhot 010080, China
Show less
文章历史+
Received
Published
Revised
2025-03-03
2026-02-28
2025-11-21
Online Date
2026-05-30
PDF (11775K)
摘要
语义同步定位与地图构建(simultaneous localization and mapping, SLAM)能够帮助移动机器人实现对未知环境更高层次的语义信息感知,已经成为解决移动机器人适应未来发展的关键技术之一。针对移动机器人在弱纹理、存在动态物体的未知境下环境感知能力弱的问题,提出了一种基于稀疏邻接注意力的图像分割算法(SNA-Seg)。首先,设计了基于稀疏邻接窗口自注意机制的全景分割网络结构,在不增加计算复杂度的前提下,充分挖掘全局信息,增强了网络对边缘细节的分割效果。其次,选取Cityscapes数据集作为训练与测试数据集,采用全景质量(panoptic quality,PQ)、分割质量(segmentation quality,SQ)和平均交并比(mIoU)等指标对算法性能进行了评估,并且采集本地视觉图像数据验证了算法的实际有效性。实验结果表明,SNA-Seg算法与基于滑窗注意力(Swin)和邻接注意力(NA)图像分割算法比较,各项评价指标均有不同程度的提升,其中mIoU指标提升幅度达到了11.17%,反映了掩膜分类准确性提升最为显著;在实例分割任务中,SNA-Seg算法展现出更高的分割精度,其输出掩膜与原始图像在边缘细节和语义类别上一致性较强,分割结果更加符合真实场景的语义结构。本文方法为语义SLAM中的全景图像分割任务提供了新的技术思路。
Abstract
Semantic SLAM(semantic simultaneous localization and mapping) enables mobile robots to perceive high-level semantic information in unknown environments and has become one of the key technologies for adapting to future developments. SNA-Seg (sparse neighborhood attention-based image segmentation algorithm) was proposed to address the problem of weak environmental perception in mobile robots under unknown scenes with low texture and dynamic objects. First, a panoramic segmentation network based on a sparse neighborhood window self-attention mechanism was first constructed to capture global information effectively and enhance the segmentation of edge details without increasing computational complexity. Then, the Cityscapes dataset was selected for training and testing. The algorithm’s performance was evaluated using metrics such as PQ (panoptic quality), SQ(segmentation quality), and mIoU (mean Intersection over union).In addition, local visual image data were collected to verify the practical effectiveness of the proposed algorithm. Experimental results show that, compared with image segmentation algorithms based on Swin (sliding window attention) and NA (neighborhood attention), the proposed SNA-Seg algorithm achieves improvements across all evaluation metrics. In particular, the mIoU was improved by 11.17%, indicating a significant enhancement in mask classification accuracy. In instance segmentation tasks, the SNA-Seg algorithm demonstrates higher segmentation precision, with output masks showing stronger consistency with original images in terms of edge details and semantic categories. The segmentation results better reflect the semantic structure of real-world scenes. The proposed method provides a novel technical approach for panoramic image segmentation in semantic SLAM.
随着人工智能技术的飞速发展,移动机器人的研究与应用将向着交互性更好、智能化水平更高、自适应能力更强的方向发展。为了适应机器人未来的发展趋势,需要对其环境感知方法的相关技术进行进一步的研究,特别是其在未知环境中的同步定位与地图构建(simultaneous localization and mapping, SLAM)技术[1]。SLAM是移动机器人能够在未知环境中自主移动和实时导引的关键技术,其可以在没有环境先验信息的条件下,通过对自身携带传感器信息的感知与处理,估计自身位置的同时建立周围环境的地图[2-3]。
HeJiafeng, Chen Hongwei, Luo Ddehan. A survey on deep learning-based real-time semantic segmentation algorithms[J]. Computer Engineering and Applications, 2023, 59(8): 13-27.
[3]
KazerouniI A, FitzgeraldL, DoolyG. A survey of state-of-the-art on visual SLAM[J]. Expert Systems with Applications, 2022, 205: 1-17.
ChenMingfang, HuangLiangen, WangSen, et al. A survey on visual odometry technologies for mobile robots[J]. Journal of Agricultural Machinery, 2024, 55(3): 1-20.
MaZhenrong, WuJunjie, LiQiang, et al. Research on indoor semantic SLAM system based on deep learning[J]. Science Techno-logy and Engineering, 2021, 21(20): 8346-8351.
LiuQingyun, YangHuayang, LiuTao, et al. SLAM algorithms based on fusion of LiDAR and depth cameras[J]. Journal of Agricultural Machinery, 2023, 54(11): 29-38.
LiXiaoqian, HeWei, ZhuShiqiang, et al. A survey on SLAM methods based on environmental semantic information[J]. Journal of Engineering Sciences, 2021, 43(6): 754-767.
[12]
CrespoJ, CastilloJ C, MozosO M, et al. Semantic information for robot navigation: a survey[J]. Applied Sciences, 2020, 10(2): 497-525.
WangJie, ChenZhengwei, XuZhaosheng, et al. Inter-rows navigation method of greenhouse robot based on fusion of camera and LiDAR[J]. Transactions of the Chinese Society for Agricultural Machinery, 2023, 54(3): 32-40.
QingChen, YuJing, XiaoChuangbo, et al. Research progress on deep convolutional neural network image semantic segmentation[J]. Journal of Image and Graphics, 2020, 25(6): 1069-1090.
LiuPeigang, XueKaixin, YuanHao, et al. Point cloud semantic segmentation method based on local perception[J]. Science Technology and Engineering, 2024, 24(15):6329-6337.
[19]
LiX Y, ChenD. A survey on deep learning-based panoptic segmentation[J]. Digital Signal Processing, 2022, 120: 1-20.
ZhuQi, WangZili, ZhangJingfeng, et al. Performance analysis of semantic segmentation model based on attention mechanism[J]. Science Technology and Engineering, 2022, 22(16): 6597-6604.
[22]
YanB, NiuX J, BareB, et al. Semantic segmentation guided pixel fusion for image retargeting[J]. IEEE Transactions on Multimedia, 2020, 22(3): 676-687.
[23]
LongJ, ShelhamerE, DarrellT. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 39(4): 640-651.
[24]
ChenL C, ZhuY, PapandreouG, et al. Encoder decoder with atrous separable convolution for semantic image segmentation[C]// Proceedings of the 15th European Conference on Computer Vision. Munich: Springer, 2018: 801-818.
[25]
SunK, XiaoB, LiuD, et al. Deep high-resolution representation learning for human pose estimation[J]. arXiv Preprint, arXiv: 2601.00584,2019.
[26]
WangS, GongY, XingJ, et al. RDSNet: a new deep architecture for reciprocal object detection and instance segmentation[C]// Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2020: 12208-12215.
[27]
FuJ, LiuJ, TianH, et al. Dual attention network for scene segmentation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2019: 3146-3154.
[28]
DosovitskiyA, BeyerL, KolesnikovA, et al. An image is worth 16×16 words: transformers for image recognition at scale[C]// International Conference on Learning Representations. San Diego: ICLR, 2021: 1-10.
[29]
XieE, WangW, YuZ, et al. SegFormer: simple and efficient design for semantic segmentation with transformers[J]. Advances in Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2021, 34: 12077-12090.
[30]
XieR, PangK, BaderG D, et al. MAESTER: masked autoencoder guided segmentation at pixel resolution for accurate, self-supervised subcellular structure recognition[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 3292-3301.
[31]
LiF, ZhangH, XuH, et al. Mask DINO: towards a unified transformer-based framework for object detection and segmentation[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 3041-3050.
[32]
JainJ, LiJ, ChiuM T, et al. OneFormer: one transformer to rule universal image segmentation[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 2989-2998.
BiYangyang, ZhengYuanfan, ShiCaijuan, et al. A survey on deep learning-based image panoptic segmentation[J]. Journal of Computer Science and Exploration, 2023, 17(11): 2605-2619.
LiJia, WangMeng, HuJinsong, et al. Research on lightweight semantic segmentation model for embedded devices[J]. Science Technology and Engineering, 2023, 23(22): 9490-9496.
LiuChang, ZhangHongbo, WangYingying, et al. Semantic segmentation of remote sensing images based on improved attention mechanism[J]. Science Technology and Engineering, 2022, 22(27): 11561-11568.
[41]
CordtsM, OmranM, RamosS, et al. The cityscapes dataset[EB/OL]. [2025-11-22]. https://www.cityscapes-dataset.com.