Dim-YOLOv5n昏暗场景目标检测算法

doi:10.3778/j.issn.1002-8331.2308-0143

摘要/Abstract

摘要： 相比于正常光照场景，照明不良昏暗场景干扰因素较多，图像处理较为复杂，且现有的昏暗目标检测，存在参数量大，识别准确率低等不足。针对昏暗场景下目标检测算法中存在误检与漏检等问题，提出以YOLOv5n算法为基础进行改进的昏暗场景目标检测算法Dim-YOLOv5n。利用嵌入全维动态卷积（omni-dimensional dynamic convolution，ODConv）的轻量化主干ODConv-MobileNetV2替换主干网络，在减少计算量的同时提高检测精度。基于RepGFPN（reparameterized generalized-FPN）方法设计更加轻量高效的LigGFPN（lightweight generalized-FPN）加强特征融合网络，以提高网络特征提取能力，并在此基础上，使用GhostConv（ghost convolution）替换传统卷积，以减少模型的参数量。实验结果表明，改进后算法与原算法相比，检测精度P和召回率R分别提高了5.3个百分点和5个百分点，平均精度均值mAP0.5：0.95和mAP0.5分别提升了8.2个百分点和4.6个百分点，改进的算法在保证模型较小的同时有效提高了检测准确率。

关键词: 昏暗图像, YOLOv5n, 全维动态卷积（ODConv）, MobileNetV2, RepGFPN, GhostConv

Abstract: Compared with normal lighting scenes, there are many interference factors in poorly illuminated scenes therefore image processing is more complex, and the existing dim object detection has shortcomings such as large number of parameters and low recognition accuracy. Aiming at the problems of false detection and missing detection in the target detection algorithm in dim scenes, Dim-YOLOv5n, a target detection algorithm in dim scenes based on YOLOv5n algorithm is proposed. ODconv-MobileNetV2 with omni-dimensional dynamic convolution (ODConv) is used to replace the backbone network to reduce the amount of computation and improve the detection accuracy. A more lightweight and efficient enhanced feature fusion network LigGFPN （lightweight generalized-FPN） based on RepGFPN (reparameterized generalized-FPN) method to improve the network’s feature extraction capability is designed, and on this basis, Ghost convolution is used to replace the traditional convolution to reduce the amount of model parameters. The experimental results show that, compared with the original algorithm, the detection accuracy P and recall R of the improved algorithm are increased by 5.3 percentage points and 5 percentage points, the average accuracy mAP0.5：0.95 and mAP0.5 are increased by 8.2 percentage points and 4.6 percentage points, respectively, and the improved algorithm effectively enhances detection accuracy while ensuring a relatively smaller model size.

Key words: dim image, YOLOv5n, omni-dimensional dynamic convolution (ODConv), MobileNetV2, reparameterized generalized-FPN (RepGFPN), Ghost convolution (GhostConv)

朱晓彤, 张荣芬, 刘宇红, 孙龙. Dim-YOLOv5n昏暗场景目标检测算法[J]. 计算机工程与应用, 2024, 60(11): 173-181.

ZHU Xiaotong, ZHANG Rongfen, LIU Yuhong, SUN Long. Dim-YOLOv5n Dim Scene Object Detection Algorithm[J]. Computer Engineering and Applications, 2024, 60(11): 173-181.

参考文献

[1] REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. arXiv:1506.01497, 2015.
[2] LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]//Proceedings of the 14th European Conference on Computer Vision, 2016: 21-37.
[3] LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE International Conference on Computer Vision, 2017: 2980-2988.
[4] LIU W, REN G, YU R, et al. Image-adaptive YOLO for object detection in adverse weather conditions[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2022: 1792-1800.
[5] REDMON J, FARHADI A. YOLOv3: an incremental improvement[J]. arXiv:1804.02767, 2018.
[6] KALWAR S, PATEL D, AANEGOLA A, et al. GDIP: gated differentiable image processing for object-detection in adverse conditions[J]. arXiv:2209.14922, 2022.
[7] SASAGAWA Y, NAGAHARA H. YOLO in the dark-domain adaptation method for merging multiple models[C]//Proceedings of the 16th European Conference on Computer Vision, 2020: 345-359.
[8] JIANG Q, MAO Y, CONG R, et al. Unsupervised decomposition and correction network for low-light image enhancement[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(10): 19440-19455.
[9] CARION N, MASSA F, SYNNAEVE G, et al. End-to-end object detection with transformers[C]//Proceedings of the European Conference on Computer Vision, 2020: 213-229.
[10] LIU R, MA L, MA T, et al. Learning with nested scene modeling and cooperative architecture search for low-light vision[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 45(5): 5953-5969.
[11] HONG Y, WEI K, CHEN L, et al. Crafting object detection in very low light[C]//Proceedings of the 32nd British Machine Vision Conference, 2021.
[12] ZHANG H, HAO K, PEDRYCZ W, et al. Vision transformer with convolutions architecture search[J]. arXiv:2203.10435, 2022.
[13] CUI Z, QI G J, GU L, et al. Multitask AET with orthogonal tangent regularity for dark object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 2553-2562.
[14] 麦锦文, 李浩, 康雁. 基于特征交互结构的弱光目标检测[J/OL]. 计算机工程与应用(2023-04-03)[2023-08-04]. http://kns.cnki.net/kcms/detail/11.2127.TP.20230403.1553.022.html.
MAI J W, LI H, KANG Y. Low-light object detection based on feature interaction structure[J/OL]. Computer Engineering and Applications(2023-04-03)[2023-08-04]. http://kns.cnki.net/kcms/detail/11.2127.TP.20230403.1553.022. html.
[15] LIU Z, LIN Y, CAO Y, et al. Swin Transformer: hierarchical vision transformer using shifted windows[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 10012-10022.
[16] LI C, ZHOU A, YAO A. Omni-dimensional dynamic convolution[J]. arXiv:2209.07947, 2022.
[17] XU X, JIANG Y, CHEN W, et al. DAMO-YOLO: a report on real-time object detection design[J]. arXiv:2211.15444, 2022.
[18] HAN K, WANG Y, TIAN Q, et al. GhostNet: more features from cheap operations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 1580-1589.
[19] 涂成凤, 易安林, 姚涛, 等. 轻量化YOLOv5n的高精度垃圾检测算法[J]. 计算机工程与应用, 2023, 59(10): 187-195.
TU C F, YI A L, YAO T, et al. High-precision garbage detection algorithm of lightweight YOLOv5n[J]. Computer Engineering and Applications, 2023, 59(10): 187-195.
[20] YU G, CHANG Q, LV W, et al. PP-PicoDet: a better real-time object detector on mobile devices[J]. arXiv:2111.00902, 2021.
[21] 刘昕宇, 姜长泓, 王其铭, 等. 基于SDP图像和MobilenetV2的滚动轴承故障诊断[J]. 组合机床与自动化加工技术, 2023, 588(2): 178-182.
LIU X Y, JIANG C H, WANG Q M, et al. Fault diagnosis of rolling bearing based on SDP lmage and MobileNetV2[J]. Modular Machine Tool & Automatic Manufacturing Technique, 2023, 588(2): 178-182.
[22] LIU S, QI L, QIN H, et al. Path aggregation network for instance segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 8759-8768.
[23] JIANG Y, TAN Z, WANG J, et al. GiraffeDet: a heavy-neck paradigm for object detection[J]. arXiv:2202.04256, 2022.
[24] LOH Y P, CHAN C S. Getting to know low-light images with the exclusively dark dataset[J]. Computer Vision and Image Understanding, 2019, 178: 30-42.
[25] ZHU B, WANG J, JIANG Z, et al. Autoassign: differentiable label assignment for dense object detection[J]. arXiv:2007.
03496, 2020.
[26] TIAN Z, SHEN C, CHEN H, et al. FCOS: fully convolutional one-stage object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019: 9627-9636.
[27] WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 7464-7475.