MFE-YOLO：Multi-Scale Feature Enhancement Algorithm for Pedestrian Detection in Complex Scenes

doi:10.3778/j.issn.1002-8331.2503-0168

Abstract

Abstract: To address the issues of high miss rate and insufficient false detection suppression in pedestrian detection under complex scenes caused by multi-scale distribution, dense occlusion, and environmental noise interference, this paper proposes MFE-YOLO, a multi-scale feature-enhanced pedestrian detection algorithm. First, an efficient global feature extraction module (EGM) is designed by integrating MetaFormer architecture with depthwise separable convolution to enhance the backbone network’s capability in capturing fine-grained pedestrian features. Additionally, a hybrid multi-level feature fusion network (MLHS-FPN), driven by a mixed local feature selection mechanism (MLCA), is constructed to dynamically fuse local details and global contextual information through cross-layer gating units, thereby improving the robustness of multi-scale pedestrian representations. A shallow P2 detection head is introduced to leverage high-resolution shallow features and deep semantic information for small target detection. Finally, the Wise-inner-PIoUv2 loss function is adopted to suppress harmful gradients from low-quality samples. Experiments on the CityPersons dataset demonstrate that MFE-YOLO achieves significant improvements over the baseline model, with F1-score and AP50 increasing by 3.7 percentage points and 6.5 percentage points, respectively, while reducing model parameters by 37% and size by 35.8%. The algorithm maintains robust detection performance in more complex scenarios, as validated on the BDD100K dataset. Results indicate that MFE-YOLO exhibits high accuracy for multi-scale pedestrian detection in challenging environments.

Key words: deep learning, pedestrian detection, MetaFormer architecture, ?multi-scale feature enhancement, YOLOv8

摘要： 针对复杂场景中行人检测存在的多尺度分布、密集遮挡及噪声干扰导致的漏检率高、误检抑制不足等问题，提出一种面向复杂场景的多尺度特征增强行人检测算法MFE-YOLO。设计高效全局特征提取模块（EGM），融合MetaFormer架构与深度可分离卷积，强化主干网络对细粒度特征的捕获能力。构建混合局部特征选择机制（MLCA）驱动的混合多级特征融合网络（MLHS-FPN），通过跨层级门控单元动态融合局部细节与全局上下文信息，提升多尺度行人表征鲁棒性。同时增加浅层小目标检测头P2，结合高分辨率特征与深层语义信息。采用Wise-inner-PIoUv2损失函数抑制低质量样本梯度干扰。在CityPersons数据集上的实验表明，MFE-YOLO相较基线模型F1分数提升3.7个百分点，AP50提高6.5个百分点，模型参数量减少37%，体积压缩35.8%。在BDD100K复杂场景下的泛化测试中仍保持性能优势。结果表明，该算法在复杂场景多尺度行人检测任务中具有较高的准确性。

关键词: 深度学习, 行人检测, MetaFormer架构, 多尺度特征增强, YOLOv8

HUANG Deyi, HUANG Deqi, HUANG Haifeng, LIU Zhenhang. MFE-YOLO：Multi-Scale Feature Enhancement Algorithm for Pedestrian Detection in Complex Scenes[J]. Computer Engineering and Applications, 2026, 62(1): 124-139.

黄德意, 黄德启, 黄海峰, 刘振航. MFE-YOLO：复杂场景下多尺度特征增强行人检测算法[J]. 计算机工程与应用, 2026, 62(1): 124-139.

References

[1] World Health Organization.Global status report on road safety 2023[R].Geneva：WHO Press，2023.
[2] VIOLA P，JONES M J.Robust real-time face detection[J].International Journal of Computer Vision，2004，57（2）：137-154.
[3] DALAL N，TRIGGS B.Histograms of oriented gradients for human detection[C]//Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2005：886-893.
[4] FELZENSZWALB P F，GIRSHICK R B，MCALLESTER D，et al.Object detection with discriminatively trained part-based models[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2010，32（9）：1627-1645.
[5] GIRSHICK R，DONAHUE J，DARRELL T，et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition.New York：ACM，2014：580-587.
[6] GIRSHICK R.Fast R-CNN[C]//Proceedings of the 2015 IEEE International Conference on Computer Vision.Piscataway：IEEE，2016：1440-1448.
[7] REN S Q，HE K M，GIRSHICK R，et al.Faster R-CNN：towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2017，39（6）：1137-1149.
[8] CAI Z W，VASCONCELOS N.Cascade R-CNN：high quality object detection and instance segmentation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2021，43（5）：1483-1498.
[9] LIU W，ANGUELOV D，ERHAN D，et al.SSD：single shot multibox detector[C]//Proceedings of the European Conference on Computer Vision.Cham：Springer，2016：21-37.
[10] CHEN S F，SUN P Z，SONG Y B，et al.DiffusionDet：diffusion model for object detection[C]//Proceedings of the 2023 IEEE/CVF International Conference on Computer Vision.Piscataway：IEEE，2024：19773-19786.
[11] 张小艳，王苗.改进的YOLOv8n轻量化景区行人检测方法研究[J].计算机工程与应用，2025，61（2）：84-96.
ZHANG X Y，WANG M.Research on improved YOLOv8n light-weight pedestrian detection method in scenic spots[J].Computer Engineering and Applications，2025，61（2）：84-96.
[12] 朱玉敏，孙光灵，缪飞.基于改进YOLOv8算法的鱼眼图像下行人检测[J].计算机科学与探索，2025，19（2）：443-453.
ZHU Y M，SUN G L，MIAO F.Pedestrian detection in fisheye images based on improved YOLOv8 algorithm[J].Journal of Frontiers of Computer Science and Technology，2025，19（2）：443-453.
[13] HE J Z，WANG J，HAN Z Y，et al.Cancer detection for small-size and ambiguous tumors based on semantic FPN and transformer[J].PLoS One，2023，18（2）：e0275194.
[14] LIU Z Y，TAN Y C，HE Q，et al.SwinNet：swin transformer drives edge-aware RGB-D and RGB-T salient object detection[J].IEEE Transactions on Circuits and Systems for Video Technology，2022，32（7）：4486-4497.
[15] WANG C C，HE W，NIE Y，et al.Gold-YOLO：efficient object detector via gather-and-distribute mechanism[C]//Advances in Neural Information Processing Systems，2024：51094-51112.
[16] YU W H，SI C Y，ZHOU P，et al.MetaFormer baselines for vision[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2024，46（2）：896-912.
[17] SANDLER M，HOWARD A，ZHU M L，et al.MobileNetV2：inverted residuals and linear bottlenecks[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2018：4510-4520.
[18] CHEN Y F，ZHANG C Y，CHEN B，et al.Accurate leukocyte detection based on deformable-DETR and multi-level feature fusion for aiding diagnosis of blood diseases[J].Computers in Biology and Medicine，2024，170：107917.
[19] 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.
[20] WAN D H，LU R S，SHEN S Y，et al.Mixed local channel attention for object detection[J].Engineering Applications of Artificial Intelligence，2023，123：106442.
[21] LIU C，WANG K G，LI Q，et al.Powerful-IoU：more straightforward and faster bounding box regression loss with a nonmonotonic focusing mechanism[J].Neural Networks，2024，170：276-284.
[22] ZHANG H，XU C，ZHANG S J.Inner-IoU：more effective intersection over union loss with auxiliary bounding box[EB/OL].[2025-03-05].https：//arxiv.org/abs/2311.02877.
[23] TONG Z J，CHEN Y H，XU Z W，et al.Wise-IoU：bounding box regression loss with dynamic focusing mechanism[EB/OL].[2025-03-05].https：//arxiv.org/abs/2301.10051.
[24] FENG C J，ZHONG Y J，GAO Y，et al.TOOD：task-aligned one-stage object detection[C]//Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision.Piscataway：IEEE，2022：3490-3499.
[25] HU J，ZHOU Y Q，WANG H，et al.Research on deep learning detection model for pedestrian objects in complex scenes based on improved YOLOv7[J].Sensors，2024，24（21）：6922.
[26] WANG Z Y，LI C，XU H Y，et al.Mamba YOLO：a simple baseline for object detection with state space model[EB/OL].[2025-03-05].https：//arxiv.org/abs/2406.05835.
[27] HU M，ZHANG Y R，JIAO T，et al.An enhanced feature-fusion network for small-scale pedestrian detection on edge devices[J].Sensors，2024，24（22）：7308.
[28] LI Y Y，HU J，WEN Y，et al.Rethinking vision transformers for MobileNet size and speed[C]//Proceedings of the 2023 IEEE/CVF International Conference on Computer Vision.Piscataway：IEEE，2024：16843-16854.
[29] HUANG S H，LU Z C，CUN X D，et al.DEIM：DETR with improved matching for fast convergence[EB/OL].[2025-03-05].https：//arxiv.org/abs/2412.04234.
[30] QIN D F，LEICHNER C，DELAKIS M，et al.MobileNetV4：universal models for the mobile ecosystem[C]//Proceedings of the European Conference on Computer Vision.Cham：Springer，2025：78-96.
[31] LIU X Y，PENG H W，ZHENG N X，et al.EfficientViT：memory efficient vision transformer with cascaded group attention[C]//Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2023：14420-14430.
[32] HOU Q B，ZHOU D Q，FENG J S.Coordinate attention for efficient mobile network design[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2021：13708-13717.
[33] LEE Y，PARK J.CenterMask：real-time anchor-free instance segmentation[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2020：13903-13912.
[34] XU W，WAN Y.ELA：efficient local attention for deep convolutional neural networks[EB/OL].[2025-03-05].https：//arxiv.org/abs/2403.01123.
[35] CAI X H，LAI Q X，WANG Y W，et al.Poly kernel inception network for remote sensing detection[C]//Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2024：27706-27716.
[36] GUO M H，LIU Z N，MU T J，et al.Beyond self-attention：external attention using two linear layers for visual tasks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2023，45（5）：5436-5447.
[37] YANG G Y，LEI J，ZHU Z K，et al.AFPN：asymptotic feature pyramid network for object detection[C]//Proceedings of the 2023 IEEE International Conference on Systems，Man，and Cybernetics.Piscataway：IEEE，2024：2184-2189.
[38] KANG M，TING C M，TING F F，et al.ASF-YOLO：a novel YOLO model with attentional scale sequence fusion for cell instance segmentation[J].Image and Vision Computing，2024，147：105057.
[39] TANG F L，XU Z X，HUANG Q M，et al.DuAT：dual-aggregation transformer network for medical image segmentation[C]//Proceedings of the Chinese Conference on Pattern Recognition and Computer Vision.Singapore：Springer，2024：343-356.
[40] DU Z W，HU Z J，ZHAO G Y，et al.Cross-layer feature pyramid transformer for small object detection in aerial images[EB/OL].[2025-03-05].https：//arxiv.org/abs/2407.19696.