Lightweight Assembly Workpiece Detection Algorithm Based on Improved YOLOv8

doi:10.16182/j.issn1004731x.joss.24-0910

Abstract

Abstract:

To address the issues of low recognition accuracy and slow detection speed with existing deep learning-based object detection algorithms for robotic automatic assembly tasks, a lightweight assembly workpiece object detection algorithm based on YOLOv8 was proposed. The PConv was introduced to improve the C2f module, anda new Faster_C2f module was designed to enhance the detection speed of the model. The SIoU loss function was employed to optimize the location prediction accuracy of the CIoU loss function and improve the localization accuracy of small targets. The high-level screening-feature fusion pyramid networks (HS-FPN) structure was used to improve the Neck part, which significantly reduced the model's computational load and increased the detection speed. By combining the iRMB module with the EMA attention mechanism, an iEMA mechanism was proposed. Experimental results have indicated that on the self-made assembly workpiece dataset, the improved algorithm increases the F1 score by 3.2%, the mAP value by 1.7%, and the FPS by 11%, while reducing the model's parameter count by 47% and the computational complexity by 32% compared with the original algorithm.

Key words: YOLOv8, attention mechanism, light weight, robotic automatic assembly, object detection

CLC Number:

TP391.41

Wu Shuheng, Liu Yongkui, Zhang Lin, Xiao Yingying, Wang Lihui. Lightweight Assembly Workpiece Detection Algorithm Based on Improved YOLOv8[J]. Journal of System Simulation, 2025, 37(12): 3099-3111.

Figures/Tables 20

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Fig. 7

Fig. 8

Fig. 9

Fig. 10

Fig. 11

Fig. 12

Table 1

Fig. 13

Fig. 14

Table 2

Table 3

Table 4

Fig. 15

Fig. 16

References 26

[1]	刘检华, 孙清超, 程晖, 等. 产品装配技术的研究现状、技术内涵及发展趋势[J]. 机械工程学报, 2018, 54(11): 1-28.
	Liu Jianhua, Sun Qingchao, Cheng Hui, et al. The State-of-the-art, Connotation and Developing Trends of the Products Assembly Technology[J]. Journal of Mechanical Engineering, 2018, 54(11): 1-28.
[2]	王耀南, 江一鸣, 姜娇, 等. 机器人感知与控制关键技术及其智能制造应用[J]. 自动化学报, 2023, 49(3): 494-513.
	Wang Yaonan, Jiang Yiming, Jiang Jiao, et al. Key Technologies of Robot Perception and Control and Its Intelligent Manufacturing Applications[J]. Acta Automatica Sinica, 2023, 49(3): 494-513.
[3]	Les Tomasz, Kruk Michal, Osowski Stanislaw. Automatic Recognition of Industrial Tools Using Artificial Intelligence Approach[J]. Expert Systems with Applications, 2013, 40(12): 4777-4784.
[4]	Paramarthalingam Arjun, Mirnalinee T T. Machine Parts Recognition and Defect Detection in Automated Assembly Systems Using Computer Vision Techniques[J]. Revista Tecnica De La Facultad De Ingenieria Universidad Del Zulia, 2016, 39(1): 71-80.
[5]	Song Rui, Li Fengming, Quan Wei, et al. Skill Learning for Robotic Assembly Based on Visual Perspectives and Force Sensing[J]. Robotics and Autonomous Systems, 2021, 135: 103651.
[6]	Wang Xi, Jaume Soriano Pinter, Liu Zhihao, et al. A Machine Learning-based Image Processing Approach for Robotic Assembly System[J]. Procedia CIRP, 2021, 104: 906-911.
[7]	Li Jing, Gu Jinan, Huang Zedong, et al. Application Research of Improved YOLO V3 Algorithm in PCB Electronic Component Detection[J]. Applied Sciences, 2019, 9(18): 3750.
[8]	Chen Chengjun, Wang Tiannuo, Li Dongnian, et al. Repetitive Assembly Action Recognition Based on Object Detection and Pose Estimation[J]. Journal of Manufacturing Systems, 2020, 55: 325-333.
[9]	Chen Wenbai, Yang Genjian, Zhang Bo, et al. Lightweight and Fast Visual Detection Method for 3C Assembly[J]. Displays, 2024, 82: 102631.
[10]	Chen Jierun, Kao S H, He Hao, et al. Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 12021-12031.
[11]	Chen Yifei, Zhang Chenyan, Chen Ben, 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.
[12]	Zhang Jiangning, Li Xiangtai, Li Jian, et al. Rethinking Mobile Block for Efficient Attention-based Models[C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2023: 1389-1400.
[13]	Ouyang Daliang, He Su, Zhang Guozhong, et al. Efficient Multi-scale Attention Module with Cross-spatial Learning[C]//ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Piscataway: IEEE, 2023: 1-5.
[14]	Zhang Haoyang, Wang Ying, Dayoub Feras, et al. VarifocalNet: An IoU-aware Dense Object Detector[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2021: 8510-8519.
[15]	Li Xiang, Wang Wenhai, Wu Lijun, et al. Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection[C]//Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2020: 21002-21012.
[16]	Yu Jiahui, Jiang Yuning, Wang Zhangyang, et al. UnitBox: An Advanced Object Detection Network[C]//Proceedings of the 24th ACM International Conference on Multimedia. New York: ACM, 2016: 516-520.
[17]	Rezatofighi H, Tsoi N, Gwak J Y, et al. Generalized Intersection Over Union: A Metric and a Loss for Bounding Box Regression[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2019: 658-666.
[18]	Zheng Zhaohui, Wang Ping, Liu Wei, et al. Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2020: 12993-13000.
[19]	Lin T Y, Dollár Piotr, Girshick R, et al. Feature Pyramid Networks for Object Detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2017: 936-944.
[20]	Liu Shu, Qi Lu, Qin Haifang, et al. Path Aggregation Network for Instance Segmentation[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 8759-8768.
[21]	Li Yanyu, Hu Ju, Wen Yang, et al. Rethinking Vision Transformers for MobileNet Size and Speed[C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2023: 16843-16854.
[22]	Zhu Lei, Wang Xinjiang, Ke Zhanghan, et al. BiFormer: Vision Transformer with Bi-level Routing Attention[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 10323-10333.
[23]	Wang C Y, Bochkovskiy A, 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.
[24]	Wang Chengcheng, He Wei, Nie Ying, et al. Gold-YOLO: Efficient Object Detector via Gather-and-distribute Mechanism[C]//Proceedings of the 37th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2023: 51094-51112.
[25]	Zhao Yian, Wenyu Lü, Xu Shangliang, et al. DETRs Beat YOLOs on Real-time Object Detection[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 16965-16974.
[26]	Selvaraju R R, Cogswell M, Das A, et al. Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization[C]//2017 IEEE International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2017: 618-626.

参数	设置
训练时长	300
优化器	AdamW
动量	0.9
学习率	0.000 769
权重衰减	0.000 5
图像大小	640×640
批量大小	16

模型	精确率/%	召回率/%	F1值/%	平均精度均值/%	参数量	千兆次浮点计算量	帧率/(帧/s)
Baseline	95.0	93.7	94.3	97.2	5.6×10⁶	19.3	78.7
+ EMA	95.0	93.8	94.3	97.1	5.6×10⁶	19.3	76.2
+ iRMB	93.6	94.1	93.8	96.9	6.6×10⁶	50.3	69.5
+ CA	95.8	93.3	94.5	96.7	5.7×10⁶	19.5	72.1
+ Biformer^[22]	92.6	94.7	93.6	95.8	6.6×10⁶	60.6	71.1
+ iEMA	95.6	95.7	95.6	97.8	5.9×10⁶	19.6	74.6

实验	Faster_C2f	SIoU	HS-FPN	iEMA	精确率/%	召回率/%	F1值/%	平均精度均值/%	参数量	千兆次浮点计算量	帧率/(帧/s)
1					88.0	97.3	92.4	96.1	11.1×10⁶	28.7	67.1
2	√				93.4	92.1	92.7	95.3	9.7×10⁶	24.4	81.4
3		√			93.7	93.4	93.5	96.3	11.1×10⁶	28.7	67.1
4			√		94.1	89.5	91.7	97.1	7.1×10⁶	24.4	75.2
5				√	93.3	92.9	93.0	96.8	11.2×10⁶	29.0	64.1
6	√	√			94.2	94.0	94.1	96.4	9.7×10⁶	24.4	81.5
7	√	√	√		95.0	93.7	94.3	97.2	5.6×10⁶	19.3	78.7
8	√	√	√	√	95.6	95.7	95.6	97.8	5.9×10⁶	19.6	74.6

模型	精确率/%	召回率/%	F1值/%	平均精度均值/%	参数量	千兆次浮点计算量	帧率/(帧/s)
YOLOv5s	87.3	94.2	90.6	94.1	7.2	16.5	73.0
YOLOv7-tiny^[23]	83.9	87.1	85.5	88.9	6.0	13.2	80.0
GOLD-YOLO^[24]	93.1	90.8	91.6	91.2	5.6	12.1	86.4
RT-DERT^[25]	95.3	92.3	93.8	93.8	42.8	130.5	19.3
YOLOv8s	88.0	97.3	92.4	96.1	11.1	28.7	67.1
Ours	95.6	95.7	95.6	97.8	5.9	19.6	74.6