ResFPN：扩增实际感受野和改进FPN的多尺度目标检测方法

doi:10.3778/j.issn.1002-8331.2401-0262

摘要/Abstract

摘要： 针对多尺度目标检测中主干网络实际感受野远远小于理论感受野，感受野分布稀疏，以及特征金字塔网络（feature pyramid network，FPN）在横向连接过程中统一通道数会丢失通道信息等影响模型性能的问题，提出一种扩增实际感受野和多特征融合改进FPN的多尺度目标检测算法ResFPN。针对主干网络实际感受野远远小于理论感受野的问题，设计了多分支膨胀卷积（multi-branch dilated convolutional，MBD）模块和多分支池化（multi-branch pooling，MBP）模块，通过学习不同尺度空间特征融合，扩增感受野。针对感受野分布稀疏问题，提出轻量级通道交互融合（channel interactive fusion，CIF）模块，通过双分支结构并在每一分支叠加不同数量深度可分离卷积学习像素间的依赖关系增强特征表示。针对FPN通过1×1卷积统一通道数会丢失通道信息的问题，尝试利用SubPixel卷积提取C5层输出特征，保留原始丰富语义信息的同时引出额外双向路径对FPN通道信息进行补充，但这可能会产生冗余信息。因此，在额外双向路径后引入全局上下文（global context，GC）模块，利用GC瓶颈转换模块进一步融合特征信息，减少信息冗余。实验表明，提出的ResFPN有效解决了感受野分布稀疏问题，并将主干网络感受野增大为原来的一倍，同时提出的改进FPN通道丢失问题的方法也在多尺度目标检测中获得了良好的性能。与典型的网络Faster R-CNN相比，大、中、小物体检测平均精度在具有挑战性的MS COCO数据集上分别提高了2.2、1.6、2.0个百分点，与其他检测器相比检测效果也有提升。

关键词: 目标检测, 卷积神经网络, 多尺度目标检测, 感受野, 特征金字塔网络（FPN）

Abstract: In view of the problems that the actual receptive field of the backbone network is much smaller than the theoretical receptive field, the sparse receptive field distribution, and the unified channel number of the feature pyramid network (FPN) in the horizontal connection process affect the performance of the model, ResFPN is proposed, which is a multi-scale object detection algorithm that expands the actual receptive field and improves the FPN by multi-feature fusion. In view of the fact that the actual receptive field of the backbone network is much smaller than the theoretical receptive field, a multi-branch dilated convolutional (MBD) module and a multi-branch pooling (MBP) module are designed to expand the receptive field by learning different scale spatial fusion. To solve the problem of sparse receptive field distribution, a lightweight channel interactive fusion (CIF) module is proposed, and the feature representation is enhanced by a two-branch structure, and the dependency relationship between separable convolution learning pixels is superimposed on each branch with different number of depths. In order to solve the problem that FPN will lose channel information through the unified channel number of 1×1 convolution, SubPixel convolution is tried to extract C5 layer features, retain the original rich semantic information and induce additional bidirectional paths to supplement the FPN channel information, but this may produce redundant information. Therefore, the global context (GC) module is introduced after the additional bidirectional path, and the GC bottleneck conversion module is used to further fuse the feature information and reduce the information redundancy. Experiments show that the proposed ResFPN effectively solves the problem of sparse receptive field distribution, and doubles the receptive field of the backbone network. Meanwhile, the proposed method to improve the FPN channel loss problem also achieves good performance in multi-scale object detection. Compared with the typical network Faster R-CNN, the average accuracy of large, medium and small object detection on the challenging MS COCO dataset is improved by 2.2, 1.6 and 2.0 percentage points, respectively, and the detection effect is also improved compared with other detectors.

Key words: object detection, convolutional neural network, multi-scale object detection, receptive field, feature pyramid network (FPN)

杨扬, 唐晓芬. ResFPN：扩增实际感受野和改进FPN的多尺度目标检测方法[J]. 计算机工程与应用, 2025, 61(10): 247-257.

YANG Yang, TANG Xiaofen. ResFPN: Multi-Scale Object Detection Algorithm for Expanding Actual Receptive Field and Improving FPN[J]. Computer Engineering and Applications, 2025, 61(10): 247-257.

参考文献

[1] 胡鹏程, 唐诗华, 张炎, 等. 决策中值滤波联合三维块匹配的遥感影像去噪[J]. 海洋测绘, 2023, 43(2): 60-64.
HU P C, TANG S H, ZHANG Y, et al. Remote sensing image denoising based on efficient decision median filtering and 3D block matching[J]. Hydrographic Surveying and Charting, 2023, 43(2): 60-64.
[2] 马妍, 古丽米拉·克孜尔别克. 图像语义分割方法在高分辨率遥感影像解译中的研究综述[J]. 计算机科学与探索, 2023, 17(7): 1526-1548.
MA Y, GULIMILA·KEZIERBIEKE. Research review of image semantic segmentation method in high-resolution remote sensing image interpretation[J]. Journal of Frontiers of Computer Science and Technology, 2023, 17(7): 1526-1548.
[3] LIN T Y, DOLLáR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017: 936-944.
[4] NAYAN A A, SAHA J, MOZUMDER A N, et al. Real time detection of small objects[J]. International Journal of Innovative Technology and Exploring Engineering, 2020, 9: 837.
[5] LIU Z M, GAO G Y, SUN L, et al. HRDNet: high-resolution detection network for small objects[C]//Proceedings of the 2021 IEEE International Conference on Multimedia and Expo. Piscataway: IEEE, 2021: 1-6.
[6] LIM J S, ASTRID M, YOON H J, et al. Small object detection using context and attention[C]//Proceedings of the 2021 International Conference on Artificial Intelligence in Information and Communication. Piscataway: IEEE, 2021: 181-186.
[7] SHEN W X, QIN P L, ZENG J C. An indoor crowd detection network framework based on feature aggregation module and hybrid attention selection module[C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision Workshop. Piscataway: IEEE, 2019: 82-90.
[8] FU K, LI J, MA L, et al. Intrinsic relationship reasoning for small object detection[J]. arXiv:2009.00833, 2020.
[9] 张华卫, 张文飞, 蒋占军, 等. 引入上下文信息和Attention Gate的GUS-YOLO遥感目标检测算法[J]. 计算机科学与探索, 2024, 18(2): 453-464.
ZHANG H W, ZHANG W F, JIANG Z J, et al. GUS-YOLO remote sensing target detection algorithm introducing context information and attention gate[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(2): 453-464.
[10] LUO W J, LI Y J, URTASUN R, et al. Understanding the effective receptive field in deep convolutional neural networks[C]//Proceedings of the 30th International Conference on Neural Information Processing Systems, 2016: 4905-4913.
[11] GAO S H, LI Z Y, HAN Q, et al. RF-next: efficient receptive field search for convolutional neural networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(3): 2984-3002.
[12] LIU Y G, YU J Z, HAN Y H. Understanding the effective receptive field in semantic image segmentation[J]. Multimedia Tools and Applications, 2018, 77(17): 22159-22171.
[13] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 770-778.
[14] LUO Y H, CAO X, ZHANG J T, et al. CE-FPN: enhancing channel information for object detection[J]. Multimedia Tools and Applications, 2022, 81(21): 30685-30704.
[15] SHI W Z, CABALLERO J, HUSZáR F, et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 1874-1883.
[16] CAO Y, XU J R, LIN S, et al. GCNet: non-local networks meet squeeze-excitation networks and beyond[C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision Workshop. Piscataway: IEEE, 2019: 1971-1980.
[17] LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 8759-8768.
[18] LIU S T, HUANG D, WANG Y H. Learning spatial fusion for single-shot object detection[J]. arXiv:1911.09516, 2019.
[19] GHIASI G, LIN T Y, LE Q V. NAS-FPN: learning scalable feature pyramid architecture for object detection[C]//Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019: 7029-7038.
[20] ZHANG Z N, ZHANG L, WANG Y, et al. Bidirectional parallel feature pyramid network for object detection[J]. IEEE Access, 2022, 10: 49422-49432.
[21] GUO C, FAN B, ZHANG Q, et al. AugFPN: improving multi-scale feature learning for object detection[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 12595-12604.
[22] QIAO S Y, CHEN L C, YUILLE A. DetectoRS: detecting objects with recursive feature pyramid and switchable atrous convolution[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 10208-10219.
[23] HOWARD A G, ZHU M L, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[J]. arXiv:1704.04861, 2017.
[24] WANG X L, GIRSHICK R, GUPTA A, et al. Non-local neural networks[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 7794-7803.
[25] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 7132-7141.
[26] WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]//Proceedings of the 15th European Conference on Computer Vision. Cham: Springer, 2018: 3-19.
[27] LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway: IEEE, 2017: 2999-3007.
[28] LI X, WANG W H, WU L J, et al. Generalized focal loss: learning qualified and distributed bounding boxes for dense object detection[C]//Advances in Neural Information Processing Systems 33, 2020: 21002-21012.
[29] CAI Z W, VASCONCELOS N. Cascade R-CNN: delving into high quality object detection[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 6154-6162.
[30] CARION N, MASSA F, SYNNAEVE G, et al. End-to-end object detection with transformers[C]//Proceedings of the 15th European Conference on Computer Vision. Cham: Springer, 2020: 213-229.
[31] HUANG J, CHEN Z X, JONATHAN WU Q M, et al. CATFPN: adaptive feature pyramid with scale-wise concatenation and self-attention[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(12): 8142-8152.