[1]张超宇,王应彪,颜旭,等.基于改进ShuffleNet V2网络的核桃破壳物料壳仁分类识别方法[J].江苏农业学报,2023,(04):1015-1025.[doi:doi:10.3969/j.issn.1000-4440.2023.04.011]
　ZHANG Chao-yu,WANG Ying-biao,YAN Xu,et al.Classification and recognition method of walnut shell and kernel based on improved ShuffleNet V2[J].,2023,(04):1015-1025.[doi:doi:10.3969/j.issn.1000-4440.2023.04.011]
点击复制

基于改进ShuffleNet V2网络的核桃破壳物料壳仁分类识别方法()
分享到：

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2023年04期
页码:
1015-1025
栏目:
农业信息工程
出版日期:
2023-08-30

文章信息/Info

Title:
Classification and recognition method of walnut shell and kernel based on improved ShuffleNet V2
作者:
张超宇; 王应彪; 颜旭; 王周梅; 李九峰; 刘梦迪; 周丹
（西南林业大学机械与交通学院，云南昆明650224）
Author(s):
ZHANG Chao-yu; WANG Ying-biao; YAN Xu; WANG Zhou-mei; LI Jiu-feng; LIU Meng-di; ZHOU Dan
(College of Mechanics and Transportation， Southwest Forestry University， Kunming 650224， China)
关键词:
C-ShuffleNet模型; ShuffleNet V2模型; 深纹核桃; 分类识别; 轻量化网络
Keywords:
C-ShuffleNet model; ShuffleNet V2 model; deep-grained walnut; classification and recognition; lightweight network
分类号:
TP391.41
DOI:
doi:10.3969/j.issn.1000-4440.2023.04.011
文献标志码:
A
摘要:
核桃破壳后壳仁物料的高效无损分类检测是云南深纹核桃加工的关键技术缺口。本研究首先使用单独的下采样模块、网络浅层不使用深度卷积及网络深层使用H-Swish激活函数替代ReLu激活函数等优化策略，将ShuffleNet V2-0.5网络模型改进为C-ShuffleNet网络模型，实现轻量化的核桃破壳物料壳仁分类检测；然后，用核桃破壳物料壳仁数据集对改进前后的模型进行训练，进而对改进后的模型进行评估与检验；最后，将改进后的模型C-ShuffleNet与AlexNet、ResNet、EfficientNet、MobileNet等经典分类网络模型进行综合性能比较。结果表明，改进后的C-ShuffleNet模型大小比改进前的ShuffleNet V2-0.5压缩了89%，测试集准确率达到9834%，比改进前提高了128个百分点，模型推理速度两者相差不大；与AlexNet、ResNet、EfficientNet、MobileNet等模型相比，C-ShuffleNet模型不但能保证较高的识别准确率，同时所占内存空间较小，识别时间更短，更加适合在嵌入式平台上开发应用。本研究结果为深纹核桃破壳物料壳仁自动化分类检测平台的开发提供了算法支持。
Abstract:
The efficient and non-destructive classification and detection of shell and kernel materials after walnut shell breaking is a key technical gap in the processing of deep-grained walnuts in Yunnan. In this study, the ShuffleNet V2-0.5 network model was improved to C-ShuffleNet network model by using optimization strategies such as a separate down-sampling module, no deep convolution in the shallow layer of the network, and H-Swish activation function instead of ReLu activation function in the deep layer of the network to realize lightweight classification detection of walnut shell and kernel. Then, the walnut shell and kernel data set was trained with ShuffleNet V2-0.5 model and C-ShuffleNet model, and C-ShuffleNet model was evaluated and tested accordingly. Finally, C-ShuffleNet model was compared with classical classification network models such as AlexNet, ResNet, EfficientNet and MobileNet. The results showed that the size of the improved C-ShuffleNet model was 8.9 % smaller than that of the ShuffleNet V2-0.5, and the accuracy of the test set reached 98.34%, which was 1.28 percentage points higher than that before the improvement. There was no significant difference in the inference speed between the two models. Compared with AlexNet, ResNet, EfficientNet and MobileNet, the C-ShuffleNet model could not only ensure higher recognition accuracy, but also occupy less memory space and shorter recognition time, which was more suitable for the development and application on embedded platforms. The results of this study provide algorithm support for the development of automatic classification and detection platform for deep-grained walnut shell-breaking materials.

参考文献/References:

[1]李娅,余红红. 基于全产业链视角的云南省核桃产业国内竞争力分析[J].林业经济问题,2018，38(5)：38-43，104.
[2]宁德鲁,王卫斌,贺熙勇,等. 云南坚果产业发展状况及SWOT分析[J].西部林业科学,2019，48(2)：8-13.
[3]刘星星,王烁烁,徐丽明,等. 基于OpenCV的动态葡萄干色泽实时识别[J].农业工程学报,2019,35(23):177-184.
[4]李颀,王康,强华，等. 基于颜色和纹理特征的异常玉米种穗分类识别方法[J].江苏农业学报,2020,36(1):24-31.
[5]甘雨,郭庆文,王春桃,等. 基于改进EfficientNet模型的作物害虫识别[J].农业工程学报,2022,38(1):203-211.
[6]KOKLU M , CINAR I, TASPINAR Y S . Classification of rice varieties with deep learning methods[J]. Computers and Electronics in Agriculture, 2021, 187: 106285.
[7]SUBEESH A, BHOLE S, SINGH K, et al. Deep convolutional neural network models for weed detection in polyhouse grown bell peppers[J]. Artificial Intelligence in Agriculture, 2022, 6:47-54.
[8]牛学德,高丙朋,南新元,等. 基于改进DenseNet卷积神经网络的番茄叶片病害检测[J].江苏农业学报,2022,38(1)：129-134.
[9]李文宝,曹成茂,张金炎,等. 基于深度学习的山核桃破壳物料分类识别[J].食品与机械,2021,37(9):133-138，185.
[10]赵腾飞,胡国玉,周建平,等. 卷积神经网络算法在核桃仁分类中的研究[J].中国农机化学报,2022,43(6):181-189.
[11]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.
[12]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.
[13]蔡建,周军,史建新,等. 基于机器视觉的核桃仁分级方法[J].江苏农业科学,2018,46(5):158-161.
[14]SANDLER M, HOWARD A, ZHU M, et al. MobileNet V2: Inverted residuals and linear bottlenecks[C].Salt Lake City：IEEE,2018: 4510-4520.
[15]ZHANG X, ZHOU X, LIN M, et al. ShuffleNet: An extremely efficient convolutional neural network for mobile devices[C]. Salt Lake City:CVF,2018：6848-6856.
[16]MA N, ZHANG X, ZHENG H T, et al. ShuffleNet V2: Practical guidelines for efficient cnn architecture design[C].Munich:LNCS,2018: 116-131.
[17]LIU Z, MAO H, WU C Y, et al. A convnet for the 2020s[C]. New Orleans:CVF,2022: 11976-11986.
[18]TAN M, LE Q V. EfficientNet V2: Smaller models and faster training[C]. Graz, Austria:PMLR, 2021: 10096-10106.
[19]HOWARD A, SANDLER M, CHU G, et al. Searching for mobilenet V3[C].Long Beach:CVF, 2019: 1314-1324.
[20]GAO H, YU S, ZHUANG L, et al. Deep networks with stochastic depth[C]. Amsterdam:LNCS, 2016: 646-661.
[21]KRIZHEVSKY A, SUTSKEVER I, HINTON G. Imagenet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
[22]HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]. Las Vegas:IEEE,2016: 770-778.
[23]SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM: Visual explanations from deep networks via gradient-based localization[J]. International Journal of Computer Vision, 2020, 128(2):336-359.

备注/Memo

备注/Memo:
收稿日期：2022-09-24 基金项目：国家自然科学基金项目（52165038）；云南省教育厅科学研究基金项目（2022Y574）；云南省农业联合专项（202101BD070001-062）作者简介：张超宇（1998-），男，河北张家口人，硕士研究生，研究方向为图像识别与目标检测算法在农业上的应用。（E-mail）1745928386@qq.com 通讯作者：王应彪，（E-mail）wybjob@163.com
更新日期/Last Update: 2023-09-12