基于YOLOX-Tiny的长尾分布交通标志识别模型

doi:10.16182/j.issn1004731x.joss.23-0906

摘要/Abstract

摘要：

在智能驾驶领域，准确识别交通标志对行车安全具有重要意义，交通标志训练集往往服从长尾分布，这为交通标志识别带来极大难度。针对于长尾分布数据集训练出的模型在尾类上表现差的现象，提出一种基于YOLOX-Tiny的长尾分布交通标志识别模型。在TT100K_2021 （tsinghua-tencent 100K 2021）数据集基础上制作交通标志长尾数据集；从制作数据集图片数量、样本分布以及模型大小出发，选择YOLOX-Tiny作为基础模型；采用EQL v2（equalization loss v2）和FL（focal loss）作为分类损失和目标置信度损失，平衡分类器头尾差距，增强模型对目标置信度的预测；在颈部双向金字塔中引入上采样算子CARAFE、坐标注意力机制（coordinate attention，CA）和CAR-ASFF模块（CARAFE + adaptively spatial feature fusion），解决传统特征金字塔上不同层级特征图的反向传播冲突问题，提升特征重组效果，突出目标特征。研究结果表明：改进的YOLOX-Tiny模型在制作的长尾交通标志数据集上mAP₅₀和mAP_50：95分别达到了43.67%和29.98％，改进模型相比较其他几种目标检测模型具有更高的检测精度。

关键词: 长尾分布, YOLOX, 交通标志识别, 注意力机制, 特征重组, 多尺度特征融合

Abstract:

Accurate recognition of traffic signs plays an important role in the field of intelligent driving. Traffic sign training datasets with long-tail distribution increase the difficulty of traffic sign recognition. A traffic sign recognition model with long-tail distribution based on YOLOX-Tiny was proposed to improve the poor performance of the model trained on long-tail distribution datasets. A long-tail traffic sign dataset was created based on the TT100K_2021 (tsinghua-tencent 100K 2021) dataset. YOLOX-Tiny was chosen as the underlying model by considering picture numbers in datasets, sample distribution, and model size. Equalization loss v2 (EQL v2) was used as classification loss to balance the head and tail of the classifier, and focal loss(FL) was used as target confidence loss to enhance the model's prediction of target confidence. In order to solvethe backpropagation conflicts of feature graphs at different levels on the traditional feature pyramid, enhance the featurereorganization effect, and highlight target feature, up-sampling operator CARAFE, coordinate attention (CA), and CARAFE + adaptively spatial feature fusion modules (CAR-ASFF) were introduced to the neck bidirectional pyramid. The research results show that the improved YOLOX-Tiny model achieves 43.67% and 29.98% respectively in the long-tail traffic sign datasets, namely mAP₅₀ and mAP_50:95. The improved model has higher detection accuracy than other target detection models.

Key words: long-tail distribution, YOLOX, traffic sign recognition, attention mechanism, feature reorganization, multiscale feature fusion

中图分类号:

TP391.9

伍云鹏,付应雄,沈丽君等 . 基于YOLOX-Tiny的长尾分布交通标志识别模型[J]. 系统仿真学报, 2024, 36(11): 2503-2516.

Wu Yunpeng,Fu Yingxiong,Shen Lijun,et al . Traffic Sign Recognition Model with Long-Tail Distribution Based on YOLOX-Tiny[J]. Journal of System Simulation, 2024, 36(11): 2503-2516.

图/表 11

图1

图2

图3

图4

图5

图6

表1

消融实验结果

模型	mAP₅₀	mAP_50:95	$m A P 50 : 95 r$	$m A P 50 : 95 c$	$m A P 50 : 95 f$
YOLOX-Tiny	39.35	26.41	16.64	30.19	40.09
+EQL v2	40.95	27.56	17.49	31.91	39.85
+FL	40.29	27.85	17.10	32.54	40.87
+CARAFE	41.38	27.85	17.48	31.70	41.70
+CA	40.61	27.19	16.91	31.20	40.59
+CAR-ASFF	41.76	28.34	18.95	31.32	41.89
+EQL v2+FL	40.97	28.41	18.16	32.99	40.62
+EQL v2+FL+CARAFE	41.64	28.72	18.49	33.04	41.39
+EQL v2+FL+CARAFE+CA	42.25	29.19	19.6	32.66	42.19
+EQL v2+FL+CARAFE+CA+CAR-ASFF	43.67	29.98	20.69	32.56	44.12

表1

图7

表2

各模型性能对比结果

模型	Size	mAP₅₀	mAP_50:95	$m A P 50 : 95 r$	$m A P 50 : 95 c$	$m A P 50 : 95 f$
YOLOv5s	416	37.90	26.37	19.20	27.97	38.02
YOLOv7-Tiny	416	40.19	26.77	20.86	27.98	36.58
YOLOv8n	416	34.43	25.46	17.33	29.07	35.17
YOLOX-Tiny Faster RCNN_r50f_CAS Faster RCNN_r50f_SL	416 608 608	39.35 26.40 21.70	26.41 20.13 16.28	16.64 20.70 14.93	30.19 19.95 17.02	40.09 19.29 17.64
本文	416	43.67	29.98	20.69	32.56	44.12

表2

图8

图9

参考文献 31

1	Xie Qiwei, Hu Xiyuan, Ren Lei, et al. A Binocular Vision Application in IoT: Realtime Trustworthy Road Condition Detection System in Passable Area[J]. IEEE Transactions on Industrial Informatics, 2023, 19(1): 973-983.
2	伍晓晖, 田启川. 交通标志识别方法综述[J]. 计算机工程与应用, 2020, 56(10): 20-26.
	Wu Xiaohui, Tian Qichuan. Survey of Traffic Sign Recognition[J]. Computer Engineering and Applications, 2020, 56(10): 20-26.
3	Tabernik Domen, Skočaj Danijel. Deep Learning for Large-scale Traffic-sign Detection and Recognition[J]. IEEE Transactions on Intelligent Transportation Systems, 2020, 21(4): 1427-1440.
4	陈飞, 刘云鹏, 李思远. 复杂环境下的交通标志检测与识别方法综述[J]. 计算机工程与应用, 2021, 57(16): 65-73.
	Chen Fei, Liu Yunpeng, Li Siyuan. Survey of Traffic Sign Detection and Recognition Methods in Complex Environment[J]. Computer Engineering and Applications, 2021, 57(16): 65-73.
5	Jiao Licheng, Zhang Fan, Liu Fang, et al. A Survey of Deep Learning-based Object Detection[J]. IEEE Access, 2019, 7: 128837-128868.
6	罗会兰, 陈鸿坤. 基于深度学习的目标检测研究综述[J]. 电子学报, 2020, 48(6): 1230-1239.
	Luo Huilan, Chen Hongkun. Survey of Object Detection Based on Deep Learning[J]. Acta Electronica Sinica, 2020, 48(6): 1230-1239.
7	刘洪江, 王懋, 刘丽华, 等. 基于深度学习的小目标检测综述[J]. 计算机工程与科学, 2021, 43(8): 1429-1442.
	Liu Hongjiang, Wang Mao, Liu Lihua, et al. A Survey of Small Object Detection Based on Deep Learning[J]. Computer Engineering & Science, 2021, 43(8): 1429-1442.
8	史燕燕, 史殿习, 乔子腾, 等. 小样本目标检测研究综述[J]. 计算机学报, 2023, 46(8): 1753-1780.
	Shi Yanyan, Shi Dianxi, Qiao Ziteng, et al. A Survey on Recent Advances in Few-Shot Object Detection[J]. Chinese Journal of Computers, 2023, 46(8): 1753-1780.
9	Lim Kwangyong, Hong Yongwon, Choi Yeongwoo, et al. Real-time Traffic Sign Recognition Based on a General Purpose GPU and Deep-learning[J]. PLoS One, 2017, 12(3): e0173317.
10	王银, 王飞翔, 孙前来. 多尺度特征融合车辆检测方法[J]. 系统仿真学报, 2022, 34(6): 1219-1229.
	Wang Yin, Wang Feixiang, Sun Qianlai. Vehicle Detection Method Based on Multi Scale Feature Fusion[J]. Journal of System Simulation, 2022, 34(6): 1219-1229.
11	江金洪, 鲍胜利, 史文旭, 等. 基于YOLO v3算法改进的交通标志识别算法[J]. 计算机应用, 2020, 40(8): 2472-2478.
	Jiang Jinhong, Bao Shengli, Shi Wenxu, et al. Improved Traffic Sign Recognition Algorithm Based on YOLO v3 Algorithm[J]. Journal of Computer Applications, 2020, 40(8): 2472-2478.
12	廖璐明, 张伟, 王亚刚. 基于注意力机制的交通标志识别[J]. 小型微型计算机系统, 2021, 42(6): 1262-1266.
	Liao Luming, Zhang Wei, Wang Yagang. Traffic Sign Recognition Based on Attention Mechanism[J]. Journal of Chinese Computer Systems, 2021, 42(6): 1262-1266.
13	潘惠苹, 王敏琴, 张福泉. 基于优化YOLO-V4的交通标志检测识别方法[J]. 计算机科学, 2022, 49(11): 179-184.
	Pan Huiping, Wang Minqin, Zhang Fuquan. Traffic Sign Detection and Recognition Method Based on Optimized YOLO-V4[J]. Computer Science, 2022, 49(11): 179-184.
14	李祥坤, 杨争峰, 曾霞, 等. 一种面向图像识别的神经网络通用扰动生成算法[J]. 系统科学与数学, 2019, 39(12): 1944-1963.
	Li Xiangkun, Yang Zhengfeng, Zeng Xia, et al. An Approach to Generate Universal Adversarial Perturbation for Deep Neural Networks in Image Recognization[J]. Journal of Systems Science and Mathematical Sciences, 2019, 39(12): 1944-1963.
15	刘庆庆, 陈岑, 汪世界. 两个广泛相依随机变量在最值运算下的次指数性[J]. 应用数学学报, 2017, 40(2): 279-286.
	Liu Qingqing, Chen Cen, Wang Shijie. The Subexponentiality for Two Widely Dependent Random Variables Under the Operations of Minimum and Maximum[J]. Acta Mathematicae Applicatae Sinica, 2017, 40(2): 279-286.
16	Shen Li, Lin Zhouchen, Huang Qingming. Relay Backpropagation for Effective Learning of Deep Convolutional Neural Networks[C]//Computer Vision – ECCV 2016. Cham: Springer International Publishing, 2016: 467-482.
17	Drummond C, Holte R C. C4.5, Class Imbalance, and Cost Sensitivity: Why Under-sampling Beats Over-sampling[C]//ICML'2003 Workshop on Learning from Imbalanced Datasets II. Washington: ACM, 2003: 1-8.
18	Wang Tao, Li Yu, Kang Bingyi, et al. The Devil is in Classification: A Simple Framework for Long-tail Instance Segmentation[C]//Computer Vision–ECCV 2020. Cham: Springer International Publishing, 2020: 728-744.
19	Li Yu, Wang Tao, Kang Bingyi, et al. Overcoming Classifier Imbalance for Long-tail Object Detection With Balanced Group Softmax[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 10988-10997.
20	Wang Tong, Zhu Yousong, Zhao Chaoyang, et al. Adaptive Class Suppression Loss for Long-tail Object Detection[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2021: 3102-3111.
21	Tan Jingru, Lu Xin, Zhang Gang, et al. Equalization Loss v2: A New Gradient Balance Approach for Long-tailed Object Detection[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2021: 1685-1694.
22	龚安, 李中浩, 梁辰宏. 夜间多场景的邻近感知实时行人检测算法[J]. 中国图象图形学报, 2023, 28(9): 2693-2705.
	Gong An, Li Zhonghao, Liang Chenhong. NSPDet: Real-time Nearby-aware Pedestrian Detection Algorithm for Multi-scene Surveillance at Night[J]. Journal of Image and Graphics, 2023, 28(9): 2693-2705.
23	Zhu Zhe, Liang Dun, Zhang Songhai, et al. Traffic-sign Detection and Classification in the Wild[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2016: 2110-2118.
24	Lin T Y, Goyal P, Girshick R, et al. Focal Loss for Dense Object Detection[C]//2017 IEEE International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2017: 2999-3007.
25	Wang Jiaqi, Chen Kai, Xu Rui, et al. CARAFE: Content-aware ReAssembly of FEatures[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2019: 3007-3016.
26	Hou Qibin, Zhou Daquan, Feng Jiashi. Coordinate Attention for Efficient Mobile Network Design[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2021: 13708-13717.
27	冷坤, 秦伦明, 王悉. 基于CA-ASFF-YOLOv4的交通标志识别研究[J]. 计算机工程与应用, 2023, 59(17): 169-177.
	Leng Kun, Qin Lunming, Wang Xi. Research on Traffic Sign Recognition Based on CA-ASFF-YOLOv4[J]. Computer Engineering and Applications, 2023, 59(17): 169-177.
28	Wang C Y, Bochkovskiy Alexey, Liao Hongyuan. YOLOv7: Trainable Bag-of-freebies Sets New State-of-the-art for Real-time Object Detectors[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 7464-7475.
29	Ren Shaoqing, He Kaiming, 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.
30	Wang Jiaqi, Zhang Wenwei, Zhang Yuhang, et al. Seesaw Loss for Long-tailed Instance Segmentation[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2021: 9690-9699.
31	Tan Jingru, Wang Changbao, Li Buyu, et al. Equalization Loss for Long-tailed Object Recognition[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 11659-11668.

[1]	李超, 李贾宝, 丁才昌, 叶志伟, 左方威. 基于DRL的边缘监控任务卸载与资源分配算法[J]. 系统仿真学报, 2024, 36(9): 2113-2126.
[2]	刘沛津, 付雪峰, 孙浩峰, 何林, 刘淑婕. 一种融合CNN与Transformer的高鲁棒性目标跟踪算法[J]. 系统仿真学报, 2024, 36(8): 1854-1868.
[3]	路阳, 刘鹏飞, 许思源, 刘启旺, 顾福谦, 王鹏. 改进注意力机制嵌入PR-Net模型的水稻病害识别仿真[J]. 系统仿真学报, 2024, 36(6): 1322-1333.
[4]	刘金辉, 陈孟元, 韩朋朋, 陈何宝, 张玉坤. 面向移动机器人大视角运动的图神经网络视觉SLAM算法[J]. 系统仿真学报, 2024, 36(5): 1043-1060.
[5]	张稀柳, 张晓玲, 何敏军. 基于改进YOLOX-s的车辆检测方法研究[J]. 系统仿真学报, 2024, 36(2): 487-496.
[6]	李晨, 何明, 董晨, 李伟. 基于余弦相似性的定向注意力行为识别模型[J]. 系统仿真学报, 2024, 36(1): 67-82.
[7]	董庆庆, 吴昊, 钱文华, 孔凤玲. 基于交叉细化和循环注意力的RGB-D显著性目标检测[J]. 系统仿真学报, 2023, 35(9): 1931-1947.
[8]	于豪, 蒋锦霞, 赖晓翰, 梅峰, 王庆. 基于自适应感受野的电力设备表面缺陷检测方法[J]. 系统仿真学报, 2023, 35(7): 1572-1580.
[9]	魏赟, 罗琦, 赵迎志. 基于自适应融合和注意力细化的语义分割模型[J]. 系统仿真学报, 2023, 35(6): 1226-1234.
[10]	石鼎, 燕雪峰, 宫丽娜, 张静宣, 关东海, 魏明强. 强化学习驱动的海战场多智能体协同作战仿真算法[J]. 系统仿真学报, 2023, 35(4): 786-796.
[11]	向南, 王璐, 贾崇柳, 蹇越谋, 马小霞. 改进YOLO的遮挡行人检测仿真[J]. 系统仿真学报, 2023, 35(2): 286-299.
[12]	孙红, 张玉香, 凌岳览. 基于损失提取反馈注意网络的图像超分辨率重建研究[J]. 系统仿真学报, 2023, 35(2): 308-317.
[13]	赵为平, 陈雨, 项松, 刘远强, 王超越. 基于改进的DeepLabv3+图像语义分割算法研究[J]. 系统仿真学报, 2023, 35(11): 2333-2344.
[14]	金炜东, 张述礼, 唐鹏, 张曼. 基于稠密残差块与通道像素注意力的图像去雾网络[J]. 系统仿真学报, 2022, 34(8): 1663-1673.
[15]	邱俊杰, 郑红, 程云辉. 基于多尺度LSTM预测模型研究[J]. 系统仿真学报, 2022, 34(7): 1593-1604.