基于边缘计算的轻量化识别方法

doi:10.19678/j.issn.1000-3428.0067836

摘要/Abstract

摘要： “新零售”模式的出现,对传统零售业的转型以及提升用户消费体验具有重要意义。然而,目前市场上的无人水果秤大多存在一些问题,如识别率低、模型结构复杂、部署困难、模型推理实时性差等。针对这些问题,提出一种基于边缘计算的轻量化识别方法。选用MobileNext作为主干网络,引入轻量型的注意力模块CBAM改进MobileNext中的SandGlass模块。利用Ghost模块替换SandGlass模块中的标准1×1卷积,以压缩模型的参数量和运算量。在此基础上,使用迁移学习的策略搭配NAdam优化器训练改进后的MobileNext模型,进一步提高模型的识别精度。在Fruit Recognition数据集上的实验结果表明,该方法在水果识别任务中达到了98.95%的识别准确率,优于原MobileNext、MobileNetV2和EfficientNet-B0等轻量级模型。相较于原MobileNext模型,改进后的MobileNext模型识别准确率提高了1.17个百分点,参数量仅为1.775×10⁶,推理时间仅为16.5 ms。在实际的零售场景中,所提方法只需很小的参数量和运算量就能实现较好的识别效果,并能够部署在边缘设备上。

关键词: 深度学习, 注意力模块, 轻量级模型, 迁移学习, 边缘设备

Abstract: The emergence of the "new retail" model has significant implications in transforming the traditional retail industry and providing enhanced consumer experience. However, existing unmanned fruit weighing scales in the market still face several challenges, such as low recognition rates, complex model structures, deployment difficulties, and poor real-time inference capabilities. To address these issues, this study proposes a lightweight recognition method based on edge computing. First, MobileNext was selected as the backbone network. Second, the Convolutional Block Attention Module (CBAM), was introduced, as a lightweight attention module, to improve the SandGlass module in MobileNext. Subsequently, the Ghost module was utilized to replace the standard 1×1 convolution in the SandGlass module, thereby reducing the number of model parameters and computational complexity. Finally, the improved MobileNext model was trained using a transfer learning strategy combined with Nesterov-accelerated Adaptive moment estimation (NAdam) optimizer to further enhance recognition accuracy. In experiments on the Fruit Recognition dataset, the proposed recognition method achieved a recognition accuracy of 98.95%, outperforming lightweight models, such as the original MobileNext model, MobileNetV2, and EfficientNet-B0. Compared to the original MobileNext model, the improved MobileNext model achieved a 1.17 percentage points increase in recognition accuracy with only 1.775 million parameters and an inference time of only 16.5 ms. In practical retail scenarios, this method requires minimal parameters and computational resources to achieve satisfactory recognition performance and has been successfully deployed on edge devices.

Key words: deep learning, attention module, lightweight model, transfer learning, edge devices

中图分类号:

TP391

李致金, 汤佳辉, 闫金凤. 基于边缘计算的轻量化识别方法[J]. 计算机工程, 2024, 50(6): 287-295.

LI Zhijin, TANG Jiahui, YAN Jinfeng. Lightweight Recognition Method Based on Edge Computing[J]. Computer Engineering, 2024, 50(6): 287-295.

https://www.ecice06.com/CN/Y2024/V50/I6/287

参考文献

[1] 邹谜,伍世虔,王欣.一种用于机器人水果采摘的快速识别方法[J].农机化研究, 2019, 41(1):206-210, 252. ZOU M, WU S Q, WANG X. A fast matching recognition method for fruit picking[J]. Journal of Agricultural Mechanization Research, 2019, 41(1):206-210, 252.(in Chinese)
[2] 余长庚,刘凯.基于小波变换与Otsu阈值去噪的脐橙识别方法[J].华南农业大学学报, 2020, 41(5):109-114. YU C G, LIU K. Navel orange recognition based on wavelet transform and Otsu threshold denoising[J]. Journal of South China Agricultural University, 2020, 41(5):109-114.(in Chinese)
[3] YU L Y, XIONG J T, FANG X Q, et al. A litchi fruit recognition method in a natural environment using RGB-D images[J]. Biosystems Engineering, 2021, 204:50-63.
[4] 洪奇峰,施伟斌,吴迪,等.深度卷积神经网络模型发展综述[J].软件导刊, 2020, 19(4):84-88. HONG Q F, SHI W B, WU D, et al. Review of the development of deep convolutional neural network model[J]. Software Guide, 2020, 19(4):84-88.(in Chinese)
[5] 盖荣丽,蔡建荣,王诗宇,等.卷积神经网络在图像识别中的应用研究综述[J].小型微型计算机系统, 2021, 42(9):1980-1984. GAI R L, CAI J R, WANG S Y, et al. Research review on image recognition based on deep learning[J]. Journal of Chinese Computer Systems, 2021, 42(9):1980-1984.(in Chinese)
[6] 季长清,高志勇,秦静,等.基于卷积神经网络的图像分类算法综述[J].计算机应用, 2022, 42(4):1044-1049. JI C Q, GAO Z Y, QIN J, et al. Review of image classification algorithms based on convolutional neural network[J]. Journal of Computer Applications, 2022, 42(4):1044-1049.(in Chinese)
[7] 王春雷,张建林,李美惠,等.结合卷积Transformer的目标跟踪算法[J].计算机工程, 2023, 49(4):281-288, 296. WANG C L, ZHANG J L, LI M H, et al. Object tracking algorithm combining convolution and Transformer[J]. Computer Engineering, 2023, 49(4):281-288, 296.(in Chinese)
[8] 石磊,张吉涛,高宇飞,等.基于Transformer与BiLSTM的网络流量入侵检测[J].计算机工程, 2023, 49(3):29-36, 57. SHI L, ZHANG J T, GAO Y F, et al. Intrusion detection of network traffic based on Transformer and BiLSTM[J]. Computer Engineering, 2023, 49(3):29-36, 57.(in Chinese)
[9] 白俊卿,韩柏迅,张丰侠.基于深度学习的无人机图像语义分割算法研究[J].计算机工程, 2023, 49(4):233-239. BAI J Q, HAN B X, ZHANG F X. Deep learning-based UAV image semantic segmentation algorithm research[J]. Computer Engineering, 2023, 49(4):233-239.(in Chinese)
[10] 王朕,李豪,严冬梅,等.基于改进YOLOv5的路面病害检测模型[J].计算机工程, 2023, 49(2):15-23. WANG Z, LI H, YAN D M, et al. Pavement disease detection model based on improved YOLOv5[J]. Computer Engineering, 2023, 49(2):15-23.(in Chinese)
[11] 朱玲.基于K-means聚类算法分类的水果等级识别与应用[J].农机化研究, 2020, 42(8):46-50. ZHU L. Recognition and application of fruit classification based on K-means clustering algorithms[J]. Journal of Agricultural Mechanization Research, 2020, 42(8):46-50.(in Chinese)
[12] 张继成,李德顺.基于深度残差学习的成熟草莓识别方法[J].中国农机化学报, 2022, 43(2):136-142. ZHANG J C, LI D S. Ripe strawberry recognition method based on deep residual learning[J]. Journal of Chinese Agricultural Mechanization, 2022, 43(2):136-142.(in Chinese)
[13] LI Z B, LI F, ZHU L, et al. Vegetable recognition and classification based on improved VGG deep learning network model[J]. International Journal of Computational Intelligence Systems, 2020, 13(1):559-564.
[14] DUONG L T, NGUYEN P T, DI S C, et al. Automated fruit recognition using EfficientNet and MixNet[J]. Computers and Electronics in Agriculture, 2020, 171:105326.
[15] ZHOU D Q, HOU Q B, CHEN Y P, et al. Rethinking bottleneck structure for efficient mobile network design[C]//Proceedings of European Conference on Computer Vision. Berlin,Germany:Springer,2020:680-697.
[16] SANDLER M, HOWARD A, ZHU M L, et al. MobileNetV2:inverted residuals and linear bottlenecks[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D.C.,USA:IEEE Press,2018:4510-4520.
[17] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Washington D.C.,USA:IEEE Press,2016:13-25.
[18] CHOLLET F. Xception:deep learning with depthwise separable convolutions[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Washington D.C.,USA:IEEE Press,2017:122-135.
[19] WOO S, PARK J, LEE J Y, et al. CBAM:convolutional block attention module[C]//Proceedings of European Conference on Computer Vision. Berlin,Germany:Springer,2018:3-19.
[20] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D.C.,USA:IEEE Press,2018:7132-7141.
[21] HAN K, WANG Y H, TIAN Q, et al. GhostNet:more features from cheap operations[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D.C.,USA:IEEE Press,2020:1580-1589.
[22] TAN C Q, SUN F C, KONG T, et al. A survey on deep transfer learning[EB/OL].[2023-05-05].https://arxiv.org/abs/1808.01974.
[23] RUDER S. An overview of gradient descent optimization algorithms[EB/OL].[2023-05-05].https://arxiv.org/abs/1609.04747.
[24] HUSSAIN I, HE Q H, CHEN Z L, et al. Fruit Recognition dataset (V 1.0)[EB/OL].[2023-05-05].https://doi.org/10.5281/zenodo.1310165.
[25] 周志华.机器学习[M].北京:清华大学出版社, 2016. ZHOU Z H. Machine learning[M]. Beijing:Tsinghua University Press, 2016.(in Chinese)
[26] SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM:visual explanations from deep networks via gradient-based localization[C]//Proceedings of IEEE International Conference on Computer Vision. Washington D.C.,USA:IEEE Press,2017:618-626.

选择文件类型/文献管理软件名称

选择包含的内容