基于改进MobileNetV3—Small的甘薯外部品质分类方法#br#

doi:10.13733/j.jcam.issn.2095-5553.2025.04.030

摘要/Abstract

摘要： 传统图像处理技术依靠人工提取特征，费时费力且难以提取到准确的特征。为准确实现对甘薯发芽、霉腐、损伤和正常品质的分类，提出一种改进的MobileNetV3—Small（M3S）分类方法。使用高效通道注意力（ECA）模块替换M3S中的压缩激励（SE）模块，构建ECA—M3S模型结构；基于迁移学习训练模型，并对比不同学习率组合的训练效果；测试甘薯品质分类模型的性能，同时和多种模型进行对比，并使用Flask设计网页界面展示测试结果。结果表明，初始学习率为0.01，学习率衰减速率为0.5时,模型整体性能最优，验证准确率为92.82%，训练损失为0.049 2；和其他10种不同复杂度的模型进行对比，该模型对4类甘薯品质的召回率均高于90%，测试平均准确率为92.43%，仅比最高的模型低0.79%，比未改进的M3S高3.59%，且模型尺寸仅为4.18 MB，仅比尺寸最小的SqueezeNet模型大1.34 MB。

关键词: 甘薯, 外部品质, MobileNetV3—Small, 高效通道注意力, 迁移学习

Abstract: Traditional image processing technology relies on manual feature extraction, which is time‑consuming, labor‑intensive and difficult to extract accurate features. In order to accurately realize the classification of sweet potato germination, mildew, damage and normal quality, an improved MobileNetV3—Small (M3S) online classification method is proposed. The efficient channel attention (ECA) module was used to replace the squeeze and excitation (SE) module in M3S, the ECA—M3S model structure was constructed. The model was trained based on transfer learning, and the training effects of different learning rate combinations were compared. The performance of the trained sweet potato quality classification model was tested and compared with a variety of models, and the web interface was designed by Flask to display the test results. The results show that the overall performance of the model is the best when the initial learning rate is 0.01 and the learning rate attenuation rate is 0.5, the verification accuracy is 92.82%, and the training loss is 0.049 2. Compared with 10 other models of different complexity, the recall rate of the four types of sweet potato quality was higher than 90%, the average test accuracy was 92.43%, only 0.79% lower than the highest, 3.59% higher than the unimproved M3S, and the model size was only 4.18 MB, only 1.34 MB larger than the smallest SqueezeNet model.

Key words: sweet potato, external quality, MobileNetV3—Small, efficient channel attention, transfer learning

中图分类号:

S531
TP391.4

马瑞峻, 丁世春, 陈瑜. 基于改进MobileNetV3—Small的甘薯外部品质分类方法#br#[J]. 中国农机化学报, 2025, 46(4): 211-217.

Ma Ruijun, Ding Shichun, Chen Yu. External quality classification method of sweet potato based on improved MobileNetV3—Small[J]. Journal of Chinese Agricultural Mechanization, 2025, 46(4): 211-217.

参考文献

[ 1 ] 赵军, 田海韬. 利用机器视觉检测马铃薯外部品质方法综述[J]. 图学学报, 2017, 38(3): 382-387.
Zhao Jun, Tian Haitao. The applications of potato external quality detection using machine vision [J]. Journal of Graphics, 2017, 38(3): 382-387.
[ 2 ] 周竹, 黄懿, 李小昱, 等. 基于机器视觉的马铃薯自动分级方法[J]. 农业工程学报, 2012, 28(7): 178-183.
Zhou Zhu, Huang Yi, Li Xiaoyu, et al. Automatic detecting and grading method of potatoes based on machine vision [J]. Transactions of the Chinese Society of Agricultural Engineering, 2012, 28(7): 178-183.
[ 3 ] 李玉华, 李天华, 牛子孺, 等. 基于色饱和度三维几何特征的马铃薯芽眼识别[J]. 农业工程学报, 2018, 34(24): 158-164.
Li Yuhua, Li Tianhua, Niu Ziru, et al. Potato bud eyes recognition based on three‑dimensional geometric features of color saturation [J]. Transactions of the Chinese Society of Agricultural Engineering, 2018, 34(24): 158-164.
[ 4 ] Thien P Q, Liou N S. The development of on‑line surface defect detection system for jujubes based on hyperspectral images [J]. Computers and Electronics in Agriculture, 2022, 194: 106743.
[ 5 ] Arshaghi A, Ashourin M, Ghabeli L. Detection and classification of potato diseases potato using a new convolution neural network architecture [J]. Traitement du Signal, 2021, 38(6): 1783-1791.
[ 6 ] Wang C L, Xiao Z F. Potato surface defect detection based on deep transfer learning [J]. Agriculture, 2021, 11(9): 863.
[ 7 ] Lecun Y, Bottou L, Bengio Y, et al. Gradient‑based learning applied to document recognition [J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.
[ 8 ] Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks [J]. Communications of the ACM, 2017, 60(6): 84-90.
[ 9 ] Simonyan K, Zisserman A. Very deep convolutional networks for large‑scale image recognition [J]. arXiv preprint arXiv: 1409. 1556, 2014.
[10] He K M, Zhang X Y, Ren S Q, et al. Deep residual learning for image recognition [J]. arXiv preprint arXiv: 1512. 03385, 2015.
[11] 易振通, 吴瑰, 官端正, 等. 轻量化卷积神经网络的研究综述[J]. 工业控制计算机, 2022, 35(10): 109-111, 114.
Yi Zhentong, Wu Gui, Guan Duanzheng, et al. Survey of research on lightweight convolutional neural networks [J]. Industrial Control Computer, 2022, 35(10): 109-111, 114.
[12] 杨森, 冯全, 张建华, 等. 基于轻量卷积网络的马铃薯外部缺陷无损分级[J]. 食品科学, 2021, 42(10): 284-289.
Yang Sen, Feng Quan, Zhang Jianhua, et al. Nondestructive classification of defects in potatoes based on lightweight convolutional neural network [J]. Food Science, 2021, 42(10): 284-289.
[13] Park J, Woo S, Lee J Y, et al. BAM: Bottleneck attention module [J]. arXiv preprint arXiv: 1807. 06514, 2018.
[14] 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 Hongxing, Xu Huiming, Liu Huanai. Lightweight agricultural crops pest identification model using improved ShuffleNet V2 [J]. Transactions of the Chinese Society of Agricultural Engineering, 2022, 38(11): 161-170.
[15] Woo S, Park J, Lee J Y, et al. CBAM: Convolutional block attention module [C]. The European Conference on Computer Vision (ECCV) , 2018: 3-19.
[16] Howard A, Sandler M, Chen B, et al. Searching for MobileNetV3 [C].2019 IEEE/CVF International Conference on Computer Vision, 2019: 1314-1324.
[17] Hu J, Shen L, Albanie S, et al. Squeeze‑and‑excitation networks [C]. IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7132-7141.
[18] Wang Q L, Wu B G, Zhu P F, et al. ECA—Net: Efficient channel attention for deep convolutional neural networks [J]. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 11534-11542.

[1]	王丹阳, 梁伟红, 李玉萍, 黄贵修, . 面向不平衡数据的木薯叶部病害图像识别方法[J]. 中国农机化学报, 2025, 46(3): 101-107.
[2]	刘起, 代东南, 孙孟研, 马德新, 徐阳. 基于双重注意力融合网络结构的圣女果果实表面缺陷识别[J]. 中国农机化学报, 2025, 46(3): 153-159.
[3]	沈中华, 程虎强, , 夏爱强, 李涵. 基于RGE—UNet模型的甘蔗蔗梢识别研究[J]. 中国农机化学报, 2025, 46(3): 188-194.
[4]	郭敬涛, 吕凤, 章慧婷, 杨彪, 刘大洋. 基于迁移学习与轻量化YOLOv5s的草莓目标检测方法[J]. 中国农机化学报, 2025, 46(3): 253-260.
[5]	牛潘婷, 张宝林, 潘丽杰, 郭建鹏(. 基于改进GoogLeNet的玉米叶片病害识别及其可解释性研究[J]. 中国农机化学报, 2025, 46(1): 204-212.
[6]	赵爽, 俞永强, 苗玉彬, 刘可心 . 基于改进 Mask R-CNN的青菜杂质检测研究[J]. 中国农机化学报, 2024, 45(9): 77-82.
[7]	韩鹏飞, 宋其江, 贾梦实. 基于改进轻量化 EfficientNet-V2模型的小麦种子分类[J]. 中国农机化学报, 2024, 45(9): 111-117.
[8]	宋小鹿 , 冯娟 , 梁翔宇 , 周玺兴 . 基于 YOLOv4的水下海参检测与计数算法[J]. 中国农机化学报, 2024, 45(9): 258-264.
[9]	苏斐, 王光辉, 史艳霞, 贾然, 闫银发, 祖林禄, . 融合通道注意力机制的ResUnet作物病斑分割模型[J]. 中国农机化学报, 2024, 45(8): 228-233.
[10]	崔中凯, 周进, 邸志峰, 姜伟, 李娜, 张华. 4UJ-1700甘薯杀秧机设计与试验[J]. 中国农机化学报, 2024, 45(6): 8-12.
[11]	徐亚雷, 杨红光, 潘志国, 杨然兵, , 李林, 张还. 甘薯裸苗斜插式栽植机构及电控系统研制[J]. 中国农机化学报, 2024, 45(6): 53-57.
[12]	郭宏杰, 马睿, 王佳, 赵威, 马德新. 基于卷积神经网络的苹果叶部病害图像识别研究[J]. 中国农机化学报, 2024, 45(5): 239-245.
[13]	白玉鹏, 冯毅琨, 李国厚, 赵明富, 周浩宇, 侯志松, . 基于Vision Transformer的小麦病害图像识别算法[J]. 中国农机化学报, 2024, 45(2): 267-274.
[14]	张莉, 杨明辉, 孙嘉成. 基于注意力机制和迁移学习的小样本茶叶病害识别[J]. 中国农机化学报, 2024, 45(10): 262-268.
[15]	张云鹏, 谭芊, 许鹏, 陈栋泉, 杨然兵. 基于EDEM-ADAMS的甘薯水平移栽机构优化设计[J]. 中国农机化学报, 2024, 45(1): 35-41.