Simulation Research on Appearance Detection of Ampoules Based on Lightweight Network and Model Compression

doi:10.16182/j.issn1004731x.joss.22-FZ0925

Abstract

Abstract:

Aiming at the large scale and redundant parameters of target detection network model, which result in the difficult to deploy the ampoule bottle appearance defect detection model to edge devices, an LC-Faster R-CNN defect detection algorithm based on lightweight network and model compression is proposed. MobileNet-V2 is used as the backbone, and the redundant channels in the convolutional network are trimmed by model pruning strategy. The floating-point parameters are quantized into integers through saturation truncation mapping. Knowledge distillation is used to restore the accuracy of the compressed network. Tested on the self-built ampoule appearance defect dataset, the model volume is reduced by 69.6% and the average accuracy is 89.3%. The simulation results show that the compressed target detection model can meet the requirements of the appearance detection of ampoules in practical applications.

Key words: object detection, model pruning, parameter quantization, knowledge distillation, Faster R-CNN

CLC Number:

TP391.4

Zhihao Zhu, Yan Wang, Zhicheng Ji. Simulation Research on Appearance Detection of Ampoules Based on Lightweight Network and Model Compression[J]. Journal of System Simulation, 2022, 34(12): 2575-2583.

Figures/Tables 13

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Table 1

Fig. 6

Table 2

Fig. 7

Table 3

Table 4

Table 5

Fig. 8

References 19

1	宛丽娟. 玻璃瑕疵在线检测关键技术研究及系统实现[D]. 秦皇岛: 燕山大学, 2017.
	Wan Lijuan. Study on Key Technology and System Realization for detection of Glass Defects Online[D]. Qinhuangdao: Yanshan University, 2017.
2	杨宗芳. 基于机器视觉的西林瓶缺陷检测算法及系统研究[D]. 太原: 太原科技大学, 2016.
	Yang Zongfang. Research on Algorithms and System of Vial Defects Detection Based on Machine Vision[D]. Taiyuan: Taiyuan University of Science and Technology, 2016.
3	Zhou X, Wang Y, Zhu Q, et al. A Surface Defect Detection Framework for Glass Bottle Bottom Using Visual Attention Model and Wavelet Transform[J]. IEEE Transactions on Industrial Informatics(S1551-3203), 2019, 16(4): 2189-2201.
4	张文利, 陈开臻, 刘鈺昕, 等. 基于边缘设备的轻量化小目标果实检测模型[J]. 中国农业信息, 2021, 33(1): 28-36.
	Zhang Wenli, Chen Kaizhen, Liu Yuxin, et al. Lightweight Small Target Fruit Detection Model Based on Edge Devices[J]. China Agricultural Informatics, 2021, 33(1): 28-36.
5	熊红林, 樊重俊, 赵珊, 等. 基于多尺度卷积神经网络的玻璃表面缺陷检测方法[J]. 计算机集成制造系统, 2020, 26(4): 900-909.
	Xiong Honglin, Fan Chongjun, Zhao Shan, et al. Glass Surface Defect Detection Method Based on Multiscale Convolution Neural Network[J]. Computer Integrated Manufacturing Systems, 2020, 26(4): 900-909.
6	陈宏彩, 任亚恒, 郝存明, 等. 基于YOLOv3的医药玻璃瓶缺陷检测方法[J]. 包装工程, 2020, 41(7): 241-246.
	Chen Hongcai, Ren Yaheng, Hao Cunming, et al. Defect Detection Method for Medical Glass Bottles Based on YOLOv3[J]. Packaging Engineering, 2020, 41(7): 241-246.
7	李康. 结合对抗自编码器与LBP编码的安瓿瓶外观缺陷检测[D]. 石家庄: 河北师范大学, 2021.
	Li Kang. Ampoule Bottle Appearance Defect Detection Based on Adversarial Auto-encoder and LBP Coding[D]. Shijiazhuang: Hebei Normal University, 2021.
8	Jin D, Xu S, Tong L, et al. End Image Defect Detection of Float Glass Based on Faster Region-Based Convolutional Neural Network[J]. Traitement Du Signal, 2020, 37(5): 807-813.
9	杨国威, 许志旺, 房臣, 等. 融合剪枝与量化的目标检测网络压缩方法[J/OL]. 计算机工程与应用 [2022-02-14]. .
008	html. Yang Guowei, Xu Zhiwang, Fang Chen, et al. Object Detection Network Compression Method Based on Pruning and Quantization[J/OL]. Computer Engineering and Applications [2022-02-14]. .
10	Ren S, He K, 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(S0162-8828), 2017, 39(6): 1137-1149. DOI: 10.1109/TPAMI.2016.2577031 .
11	Simonyan K, Zisserman A. Very Deep Convolutional Networks for Large-scale Image Recognition[J/OL]. arXiv preprint arXiv:1409.1556 [2022-02-12]. .
12	Sandler M, Howard A, Zhu M, et al. MobileNetV2: Inverted Residuals and Linear Bottlenecks[C]// Computer Vision and Pattern Recognition (CVPR). Salt Lake City, UT, USA: IEEE, 2018: 4510-4520.
13	Howard A G, Zhu M, Chen B, et al. MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications[J/OL]. Computer Vision and Pattern Recognition[2022-02-12]. .
14	He K, Zhang X, Ren S, et al. Deep Residual Learning for Image Recognition[C]// Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA: IEEE, 2016: 770-778.
15	He Y, Zhang X, Sun J. Channel Pruning for Accelerating Very Deep Neural Networks[C]// IEEE International Conference on Computer Vision (ICCV). Venice, Italy: IEEE, 2017: 1398-1406.
16	Ding X, Hao T, Tan J, et al. Resrep: Lossless Cnn Pruning via Decoupling Remembering and Forgetting[C]//IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, QC, Canada: IEEE, 2021: 4510-4520.
17	Ding X, Zhang X, Ma N, et al. Repvgg: Making Vgg-Style Convnets Great Again[C]// Computer Vision and Pattern Recognition (CVPR). Nashville, TN, USA: IEEE, 2021: 13728-13737.
18	Wu J, Leng C, Wang Y, et al. Quantized Convolutional Neural Networks for Mobile Devices[C]//Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA: IEEE, 2016: 4820-4828.
19	Hinton G, Vinyals O, Dean J. Distilling the Knowledge in a Neural Network[J/OL]. arXiv preprint arXiv:1503. 02531, 2015, 2(7)[2022-02-12]. .

参数名称		参数值
Batch_Size		8
Learning rate		0.005
Step_Size		3
Momentum		0.9

剪枝次数	mAP/%	MADD/M	FLOPs/G	params/M
量化	86.38	394.4	1.28	0.70
0	91.00	1458.3	4.64	2.30
1	90.02	985.2	3.15	1.45
2	88.72	708.3	2.27	1.00
4	86.44	433.4	1.40	0.70

Input	Operator	压缩前参数量	压缩后参数量	Output
32×128×128	Conv	0.001	0.001	32×128×128
32×128×128	BN	0	0	32×128×128
32×128×128	Bottleneck	0.003	0.002	16×128×128
16×128×128	Bottleneck Bottleneck	0.006 0.009	0.003 0.004	24×128×128
24×128×128	Bottleneck Bottleneck Bottleneck	0.010 0.015 0.015	0.005 0.007 0.003	32×64×64
32×64×64	Bottleneck	0.021	0.004	64×32×32
	Bottleneck	0.054	0.010
	Bottleneck	0.054	0.010
	Bottleneck	0.054	0.010
64×32×32	Bottleneck	0.073	0.018	96×32×32
	Bottleneck	0.118	0.020
	Bottleneck	0.118	0.007
96×32×32	Bottleneck	0.155	0.010	160×16×16
	Bottleneck	0.320	0.019
	Bottleneck	0.320	0.019
160×16×16	Block	0.526	0.080	320×16×16
320×16×16	Conv	0.410	0.410	1 280×16×16
1 280×16×16	BN	0.003	0.003	1 280×16×16
1 280×16×16	Linear	0.013	0.013	16×10

蒸馏前后	mAP
蒸馏前	86.38
蒸馏后	89.30

检测类别	精确度	漏检率	误检率
平均	89.3	12.7	10.7
安瓿瓶	100	0	0
碳化黑点	91.5	8.4	8.5
头部缺陷	81.1	24.8	18.9
裂纹	84.6	17.6	15.4