基于改进ShuffleNet V2的轻量级防风药材道地性智能识别

doi:10.12133/j.smartag.SA202304003

摘要/Abstract

摘要：

［目的/意义］ 目前，对于防风药材产地和品质的鉴别方法主要是根据其物理或化学特征，其方法需对中药材进行分离提取，存在耗时长，费用高，专业性强，技术难度大等问题，不利于推广应用。随着深度学习的不断发展，其无需人工提取特征、分类精度高等优点被广泛应用在中药材的识别之中。 ［方法］ 针对大多数卷积神经网络模型在识别防风药材时计算量大、精度低的问题，本研究提出了一种改进的ShuffleNet V2的轻量级防风道地性识别模型。在不降低网络性能的情况下调整模型架构，减少模型参数量和计算量，用沙漏残差网络（Hourglass Residual Network）代替传统残差网络，同时引入SE（Squeeze-and-Excitation）注意力机制，把具有附加信道注意力的沙漏残差网络嵌入到ShuffleNet V2中，使用SiLU激活函数替换 ReLU 激活函数，丰富局部特征学习，从而提出轻量化的中药防风道地性识别模型 Shuffle-Hourglass SE。为了验证本文所提出模型的有效性，选用VGG16、MobileNet V2、ShuffleNet V2和SqueezeNet V2四种经典网络模型进行对比实验。［结果和讨论］结果表明，本研究提出的模型Shuffle-Hourglass SE获得了最佳性能。在测试集上取得95.32%的准确率、95.28%的召回率，F₁分数达到95.27%，测试时间、模型大小为246.34 ms和3.23 M，不仅在传统CNN网络中是最优的，在轻量级网络中也具有较大优势。 ［结论］ 本研究所提出的模型在保持较高识别精度的同时占用较少的储存空间，有助于在未来的低性能终端上实现防风道地性的实时诊断。

关键词: 防风, 道地性识别, ShuffleNet V2, SE注意力机制, 沙漏残差网络, 中药材, 轻量级模型

Abstract:

[Objective] Saposhnikovia divaricata (Turcz.) Schischk is a kind of traditional Chinese medicine. Currently, the methods of identifying the origin and quality of Saposhnikovia divaricata (Turcz.) Schischk are mainly based on their physical or chemical characteristics, which is impossible to make an accurate measurement of Groundness identification. With the continuous development of deep learning, its advantages of no manual extraction and high classification accuracy are widely used in different fields, and an attention-embedded ShuffleNet V2-based model was proposed in this study to address the problems of large computation and low accuracy of most convolutional neural network models in the identification of Chinese herbal medicine Saposhnikovia divaricata (Turcz.) Schischk. [Methods] The model architecture was adjusted to reduce the number of model parameters and computation without degrading the network performance, and the traditional residual network was replaced by the Hourglass residual network, while the SE attention mechanism was introduced to embed the hourglass residual network with additional channel attention into ShuffleNet V2. The important features were enhanced and the unimportant features were weakened by controlling the size of the channel ratio to make the extracted features more directional by SE attention. The SiLU activation function was used to replace the ReLU activation function to enhance the generalization ability of the model Enriching local feature learning. Therefore, a lightweight Shuffle-Hourglass SE model was proposed. The samples of Saposhnikovia divaricata (Turcz.) Schischk used in this research were samples from the main production areas, including more than 1000 samples from five production areas in Heilongjiang, Jilin, Hebei, Gansu and Inner Mongolia. A total of 5234 images of Saposhnikovia divaricata (Turcz.) Schischk were obtained by using cell phone photography indoors under white daylight, fully taking into account the geographical distribution differences of different Saposhnikovia divaricata (Turcz.) Schischk. The data set of Saposhnikovia divaricata (Turcz.) Schischk images was expanded to 10,120 by using random flip, random crop, brightness and contrast enhancement processes. In order to verify the effectiveness of the model proposed, four classical network models, VGG16, MobileNet V2, ShuffleNet V2 and SqueezeNet V2, were selected for comparison experiments, ECA ( Efficient Channel Attention ) attention mechanism, CBAM ( Convolutional Block Attention Module ) attention mechanism and CA attention mechanism were chosen to compare with SE. All attention mechanisms were introduced into the same position in the ShuffleNet V2 model, and ReLU, H-swish and ELU activation functions were selected for contrast experiments under the condition in which other parameters unchanged. In order to explore the performance improvement of ShuffleNet V2 model by using the attention mechanism of SE module, Hourglass residual block and activation function, Shuffle-Hourglass SE model ablation experiment was carried out. Finally, loss, accuracy, precision, recall and F₁ score in test set and training set were used as evaluation indexes of model performances. [Results and Discussions] The results showed that the Shuffle-Hourglass SE model proposed achieved the best performances. An accuracy of 95.32%, recall of 95.28%, and F₁ score of 95.27% were obtained in the test set, which was 2.09%, 2.1 %, and 2.19 % higher than the ShuffleNet V2 model, respectively. The test duration and model size were 246.34 ms and 3.23 M, respectively, which were not only optimal among Traditional CNN such as VGG and Desnet,but had great advantages among lightweight networks such as MobileNet V2、SqueezeNet V2 and ShufffleNet V2. Compared with the classical convolutional network VGG, 7.41% of the accuracy was improved, 71.89% of the test duration was reduced, and 96.76% of the model size was reduced by the Shuffle-Hourglass SE model proposed in this study. Although the test duration of ShuffleNet V2 and MobileNet V2 were similar, the accuracy and speed of the Shuffle-Hourglass SE model improved, which proved its better performance. Compared with MobileNet V2, the test duration was reduced by 69.31 ms, the model size was reduced by 1.98 M, and the accuracy was increased by 10.5 %. In terms of classification accuracy, the improved network maintains higher recognition accuracy and better classification performance. [Conclusions] The model proposed in this research is able to identify the Saposhnikovia divaricata (Turcz.) Schischk originality well while maintaining high identification accuracy and consuming less storage space, which is helpful for realizing real-time identification of Saposhnikovia divaricata (Turcz.) Schischk originality in the future low performance terminals.

Key words: Saposhnikovia divaricata (Turcz.) Schischk, originality recognition, ShuffleNet V2, SE attention mechanism, hourglass residual network, traditional Chinese medicine, lightweight model

中图分类号:

R282.5

赵毓, 任艺平, 朴欣茹, 郑丹阳, 李东明. 基于改进ShuffleNet V2的轻量级防风药材道地性智能识别[J]. 智慧农业(中英文), 2023, 5(2): 104-114.

ZHAO Yu, REN Yiping, PIAO Xinru, ZHENG Danyang, LI Dongming. Lightweight Intelligent Recognition of Saposhnikovia Divaricata (Turcz.) Schischk Originality Based on Improved ShuffleNet V2[J]. Smart Agriculture, 2023, 5(2): 104-114.

图/表 15

图 1

表 1

Shuffle-Hourglass SE模型内部参数

结构层	输出尺寸	卷积核大小	步长	重复数/次	输出通道
输入	224× $224$	——	——	——	——
卷积1	112×112	3 $×$ 3	2	1	24
深度可分卷积2	56×56	3 $×$ 3	2	1	24
阶段2	28×28 28×28	——	2 1	1 1	244
阶段3	14×14 14×14	——	2 1	1 1	488
阶段4	7×7 7×7	——	2 1	1 1	976
卷积5	7×7	——	1	1	2048
平均池化	1×1	7×7	——	——	——
全连接	——	——	——	——	吉林/内蒙古/甘肃/黑龙江/河北

表 1

图 2

图3

图 4

图 5

图 6

图 7

表2

图 8

图 9

表 3

表 4

表 5

表 6

Shuffle-Hourglass SE模型消融实验结果

模型	因素				准确率/%	参数量
模型	注意力机制模块	沙漏残差块	激活函数	架构调整	准确率/%	参数量
ShuffleNet V2	$×$	$×$	$×$	$×$	93.23	4.52
	√	$×$	$×$	$×$	94.53	4.52
	$×$	√	$×$	$×$	94.12	5.56
	$×$	$×$	√	$×$	94.44	4.52
	√	√	$×$	$×$	94.92	5.56
	√	√	√	$×$	95.03	5.56
Shuffle-Hourglass SE	√	√	√	√	95.32	3.23

表 6

参考文献 34

1	程显隆, 郭晓晗, 李明华, 等. 道地性和生产规范性是中药材质量属性形成的关键[J]. 中国现代中药, 2020, 22(7): 991-995, 1155.
	CHENG X L, GUO X H, LI M H, et al. Daodi attribute and production standardization are the key to the formation of the quality attributes of Chinese materia Medica [J]. Modern Chinese medicine, 2020, 22(7): 991-995, 1155.
2	胡云飞. 基于药物分析组合技术研究凤丹药材的道地性[D]. 合肥: 安徽中医药大学, 2015.
	HU Y F. Studies on the genuineness of Feng Dan based on the combination technology of pharmaceutical analysis[D]. Hefei: Anhui University of Chinese Medicine, 2015.
3	张冬月, 吴浩善, 李思雨, 等. 电子鼻技术鉴别通关藤真伪及其道地性[J]. 中成药, 2018, 40(10): 2228-2233.
	ZHANG D Y, WU H S, LI S Y, et al. Authenticity and genuineness identification of Marsdenia tenacissima by electronic nose[J]. Chinese traditional patent medicine, 2018, 40(10): 2228-2233.
4	刘飞, 王元忠, 杨春艳, 等. 红外光谱结合判别分析对三七道地性及产地的鉴别研究[J]. 光谱学与光谱分析, 2015, 35(1): 108-112.
	LIU F, WANG Y Z, YANG C Y, et al. Study on the genuineness and producing area of Panax notoginseng based on infrared spectroscopy combined with discriminant analysis[J]. Spectroscopy and spectral analysis, 2015, 35(1): 108-112.
5	ZHANG T C, LI S, WANG Y H, et al. Research progress in the application of stable isotope and mineral element analysis in tracing the geographical origin of Chinese medicinal materials[J]. Food science and technology, 2022, 42: ID e08222.
6	DENG L. Deep learning: Methods and applications[J]. Foundations and trends® in signal processing, 2014, 7(3/4): 197-387.
7	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
8	LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[C]// 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, NJ, USA: IEEE, 2015: 3431-3440.
9	REN S Q, HE K M, 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.
10	YUAN Z W, ZHANG J. Feature extraction and image retrieval based on AlexNet[C]// Proc SPIE 10033, eighth international conference on digital image processing (ICDIP 2016), 2016, 10033: 65-69.
11	HE J, LI S, SHEN J M, et al. Facial expression recognition based on VGGNet convolutional neural network[C]// 2018 Chinese Automation Congress (CAC). Piscataway, NJ, USA: IEEE, 2019: 4146-4151.
12	AL-QIZWINI M, BARJASTEH I, AL-QASSAB H, et al. Deep learning algorithm for autonomous driving using GoogLeNet[C]// 2017 IEEE Intelligent Vehicles Symposium (IV). Piscataway, NJ, USA: IEEE, 2017: 89-96.
13	TARG S, ALMEIDA D, LYMAN K. Resnet in resnet: Generalizing residual architectures[EB/OL]. arXiv: , 2016.
14	ZHU Y, NEWSAM S. DenseNet for dense flow[C]// 2017 IEEE International Conference on Image Processing (ICIP). Piscataway, NJ, USA: IEEE, 2018: 790-794.
15	赵潇潇. 一种基于外观图像的枸杞道地性AI识别研究[D]. 天津: 天津科技大学, 2021.
	ZHAO X X. A study on the recognition of genuine AI of Lycium barbarum based on appearance image [D].Tianjin: Tianjin University of Science & Technology, 2021.
16	李东明, 汤鹏, 张丽娟, 等. 采用改进稠密连接网络的防风药材的道地性识别[J]. 农业工程学报, 2022, 38(3): 276-285.
	LI D M, TANG P, ZHANG L J, et al. Genuine identification for Saposhnikovia divaricata based on improved DenseNet[J]. Transactions of the Chinese society of agricultural engineering, 2022, 38(3): 276-285.
17	LYU B J, WU L Y, HUANGFU T R, et al. Traditional Chinese medicine recognition based on target detection[J]. Evidence-based complementary and alternative medicine, 2022, 2022: 1-9.
18	XING C Y, HUO Y Q, HUANG X J, et al. Research on image recognition technology of traditional Chinese medicine based on deep transfer learning[C]// 2020 International Conference on Artificial Intelligence and Electromechanical Automation (AIEA). Piscataway, NJ, USA: IEEE, 2020: 140-146.
19	CHOLLET F. Xception: deep learning with depthwise separable convolutions[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, NJ, USA: IEEE, 2017: 1800-1807.
20	SINHA D, EL-SHARKAWY M. Thin MobileNet: An enhanced MobileNet architecture[C]// 2019 IEEE 10th Annual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON). Piscataway, NJ, USA: IEEE, 2020: 280-285.
21	SANDLER M, HOWARD A, ZHU M L, et al. MobileNetV2: inverted residuals and linear bottlenecks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ, USA: IEEE, 2018: 4510-4520.
22	MA N N, ZHANG X Y, ZHENG H T, et al. ShuffleNet V2: Practical guidelines for efficient CNN architecture design[C]// Computer Vision - ECCV 2018: 15th European Conference, Munich, Germany, September 8-14, 2018, Proceedings, Part XIV. New York, USA: ACM, 2018: 122-138.
23	刘雪纯, 刘大铭, 常佳鑫, 等. 基于MobileNet V2迁移学习的中药材图像识别[J]. 长江信息通信, 2022, 35(7): 33-37, 44.
	LIU X C, LIU D M, CHANG J X, et al. MobileNet V2 migration learning based image recognition of Chinese herbs[J]. Changjiang information & communications, 2022, 35(7): 33-37, 44.
24	XIANG Q, WANG X D, LI R, et al. Fruit image classification based on MobileNetV2 with transfer learning technique[C]// Proceedings of the 3rd International Conference on Computer Science and Application Engineering. New York, USA: ACM, 2019: 1-7.
25	ZHU H G, ZENG H M, LIU J H, et al. Logish: A new nonlinear nonmonotonic activation function for convolutional neural network[J]. Neurocomputing, 2021, 458: 490-499.
26	TANG Z, YANG J L, LI Z, et al. Grape disease image classification based on lightweight convolution neural networks and channelwise attention[J]. Computers and electronics in agriculture, 2020, 178: ID 105735.
27	LI Y, LIU Y, CUI W G, et al. Epileptic seizure detection in EEG signals using a unified temporal-spectral squeeze-and-excitation network[J]. IEEE transactions on neural systems and rehabilitation engineering, 2020, 28(4): 782-794.
28	KARAMAN A, PACAL I, BASTURK A, et al. Robust real-time polyp detection system design based on YOLO algorithms by optimizing activation functions and hyper-parameters with artificial bee colony (ABC)[J]. Expert systems with applications, 2023, 221: ID 119741.
29	张旭, 周云成, 刘忠颖, 等. 基于改进Shufflenet V2模型的苹果叶部病害识别及应用[J]. 沈阳农业大学学报, 2022, 53(1): 110-118.
	ZHANG X, ZHOU Y C, LIU Z Y, et al. Identification and application of apple leaf diseases based on improved Shufflenet V2 model[J]. Journal of Shenyang agricultural university, 2022, 53(1): 110-118.
30	李好, 邱卫根, 张立臣. 改进ShuffleNet V2的轻量级农作物病害识别方法[J]. 计算机工程与应用, 2022, 58(12): 260-268.
	LI H, QIU W G, ZHANG L C. Improved ShuffleNet V2 for lightweight crop disease identification[J]. Computer engineering and applications, 2022, 58(12): 260-268.
31	ZUO Y M, LIU P S, TAN Y Q, et al. An attention-based lightweight residual network for plant disease recognition[C]// 2020 International Conference on Artificial Intelligence and Computer Engineering (ICAICE). Piscataway, NJ, USA: IEEE, 2021: 224-228.
32	周剑, 谢知音. 基于注意力机制的改进残差网络的柑橘分类研究[J]. 现代农业装备, 2023, 44(1): 47-55.
	ZHOU J, XIE Z Y. Research on Citrus classification based on improved residual network and attention mechanism[J]. Modern agricultural equipment, 2023, 44(1): 47-55.
33	徐增敏, 陈凯, 郭威伟, 等. 面向轻量级卷积网络的激活函数与压缩模型[J]. 计算机工程, 2022, 48(5): 242-250.
	XU Z M, CHEN K, GUO W W, et al. Activation function and compression model for lightweight convolutional network[J]. Computer engineering, 2022, 48(5): 242-250.
34	彭红星, 徐慧明, 刘华鼐. 基于改进ShuffleNet V2的轻量化农作物害虫识别模型[J]. 农业工程学报, 2022, 38(11): 161-170.
	PENG H X, XU H M, LIU H N. Lightweight agricultural crops pest identification model using improved ShuffleNet V2[J]. Transactions of the Chinese society of agricultural engineering, 2022, 38(11): 161-170.

原产地	测试集/幅	训练集/幅	图像数量/幅
甘肃	404	1617	2021
河北	398	1589	1987
黑龙江	411	1642	2053
吉林	470	1879	2349
内蒙古	342	1368	1710

模型	准确率/%	测试时长/ms	召回率/%	F₁分数/%	模型大小/M
MobileNet V2	84.82	315.65	84.91	84.46	5.21
ShuffleNet V2	93.23	324.54	93.18	93.08	4.52
VGG	87.91	876.54	88.02	87.82	99.72
Shuffle-Hourglass SE	95.32	246.34	95.28	95.27	3.23
SqueezeNet V2	62.71	161.98	62.88	62.43	6.03

注意力机制	准确率/%	召回率/%	F₁分数/%	参数量/M
ECA	93.72	93.88	93.45	4.52
SE	95.32	95.44	95.31	4.39
CA	94.64	94.75	94.49	4.98
CBAM	94.13	94.22	93.93	4.52

激活函数	准确率/%	F₁分数/%
ReLU	93.23	93.11
H-Swish	94.08	93.88
ELU	92.98	92.96
SiLU	95.32	95.04