基于果穗图像的玉米品种分类识别

doi:10.13304/j.nykjdb.2022.0633

摘要/Abstract

摘要：

优良品种对提高农业产量和收入起着关键作用，针对现有的种业安全问题，为实现玉米品种的快速识别和保护，构建一种基于玉米果穗图像的品种识别模型。将采集到的1 000张玉米果穗图像经预处理后按7∶2∶1的比例划分为训练集、验证集和测试集，并对数据集进行平移、翻转等多种数据增强处理。通过迁移学习，将预训练好的权重和参数迁移到NASNet-mobile、Xception、ResNet50V2、MobileNetV2、DenseNet121、VGG16模型进行对比，结果表明，NASNet-mobile识别性能较好，识别率达90%。不同优化算法的对比表明，优化器选择Adam模型具有更好的表现。在此基础上，对多种全连接层模块进行试验，结果表明，全连接层数量为2层、维度为256时可以得到更好的玉米果穗图像特征，最终模型在全连接层模块下的识别准确率达95%，较NASNet-mobile提升5%，实现了对玉米品种的分类识别。以上结果为玉米品种的快速精准鉴定以及种质资源保护提供了智能化技术支持。

关键词: 玉米果穗, 迁移学习, 品种识别, NASNet-mobile

Abstract:

Crop variety plays a key role in improving agricultural production and income. Aiming at the safety problems of seed industry， in order to realize the rapid recognition and protection of corn varieties， a variety recognition model based on ear image was proposed. After image preprocessing， 1 000 images of corn ears were divided into training set， validation set and test set according to the ratio of 7∶2∶1. And the data sets were enhanced by translation， flipping and other data processing. Using transfer learning technology， the pre-trained weights and parameters were transferred to NASNet-mobile， Xception， ResNet50V2， MobileNetV2， DenseNet121 and VGG16 for comparative experiments. The results showed that the performance of NASNet-mobile was best， and the recognition rate reached 90%. At the same time， different optimization algorithms were used for comparative experiments， and the result showed that Adam model performed better. Based above results， experiments were carried out under a variety of different full connection layer modules. The results showed that， when the number of full connected layers was 2 and the dimension was 256， better corn ear image features could be obtained， and the recognition accuracy of the final model under the full connection layer module reached 95%， which increased by 5% compared with NASNet-mobile. It realized the variety classification and recognition of corn ear image， which provided intelligent technical support for the rapid and accurate identification of corn varieties and the protection of germplasm resources.

Key words: corn ear, transfer learning, variety identification, NASNet-mobile

中图分类号:

S513

赵威, 马睿, 王佳, 郭宏杰, 许金普. 基于果穗图像的玉米品种分类识别[J]. 中国农业科技导报, 2023, 25(6): 97-106.

Wei ZHAO, Rui MA, Jia WANG, Hongjie GUO, Jinpu XU. Classification and Identification of Corn Varieties Based on Ear Image[J]. Journal of Agricultural Science and Technology, 2023, 25(6): 97-106.

图/表 13

图1 玉米果穗图像采集装置

Fig. 1 Image acquisition device for corn ear

表 1 玉米果穗数据集

Table 1 Corn ear dataset

类别 Category	训练集 Training set	验证集 Validation set	测试集 Test set	总计 Total	训练集增强后总计 Training set enhanced total
锦玉118 Jinyu 118	140	40	20	200	980
科诺58 Kenuo 58	140	40	20	200	980
立原296 Liyuan 296	140	40	20	200	980
铁研630 Tieyan 630	140	40	20	200	980
荟玉18 Huiyu 18	140	40	20	200	980

表2 网络模型基线测试准确率

Table 2 Baseline test accuracy of network models

测试准确率

Test accuracy

图2 基于CNN和参数迁移相结合的玉米果穗图像识别流程

Fig. 2 Corn ear image recognition process based on CNN and parameter migration

图3 改进全连接层的NASNet-mobile网络结构

Fig. 3 Improving NASNet-mobile network structure of full connection layer

图4 玉米果穗图像预处理A：原始图像；B：灰度化；C：阈值分割和二值化；D：处理后图像

Fig. 4 Image preprocessing of corn earsA： Original image； B： Graying； C： Threshold segmentation and binarization； D： Processed image

图5 不同优化算法下模型的训练曲线

Fig. 5 Training curves of models under different optimization algorithms

表3 不同全连接层模块下模型性能评估

Table 3 Model performance evaluation under different full connection layer modules

指标 Index	基线 Benchmark	Batch_size：32						Batch_size：64
指标 Index	基线 Benchmark	T=1 D=256	T=2 D=256	T=3 D=256	T=1 D=512	T=2 D=512	T=3 D=512	T=1 D=256	T=2 D=256	T=3 D=256	T=1 D=512	T=2 D=512	T=3 D=512
精准率Precision	0.908 7	0.923 1	0.956 1	0.907 9	0.916 8	0.909 4	0.928 2	0.912 5	0.920 2	0.925 9	0.918 5	0.874 1	0.923 0
召回率Recall	0.900 0	0.920 0	0.950 0	0.890 0	0.890 0	0.860 0	0.920 0	0.910 0	0.910 0	0.910 0	0.880 0	0.830 0	0.920 0
调和平均值F1-score	0.901 4	0.921 5	0.950 5	0.893 4	0.892 1	0.863 6	0.920 5	0.910 6	0.909 7	0.912 1	0.883 4	0.834 8	0.920 4
准确率Accuracy	0.900 0	0.920 0	0.950 0	0.890 0	0.890 0	0.860 0	0.920 0	0.910 0	0.910 0	0.910 0	0.880 0	0.830 0	0.920 0

图6 NASNet-mobile-maize训练与验证结果

Fig. 6 Results of NASNet-mobile-maize training and validation

表4 NASNet-mobile-maize模型下不同品种玉米果穗测试结果的评价指标

Table 4 Evaluation indexes of ear test results of different varieties of maize under NASNet-mobile-maize model

指标index	荟玉18 Huiyu 18	锦玉118 Jinyu 118	科诺58 Kenuo 58	立原296 Liyuan 296	铁研630 Tieyan 630	平均Mean
精准率Precision	0.947 3	0.833 3	1.000 0	1.000 0	1.000 0	0.956 1
召回率Recall	0.900 0	1.000 0	0.900 0	0.950 0	1.000 0	0.950 0
调和平均值F1-score	0.923 0	0.909 0	0.947 3	0.974 3	1.000 0	0.950 7
准确率Accuracy	0.900 0	1.000 0	0.900 0	0.950 0	1.000 0	0.950 0

图7 NASNet-mobile-maize测试结果的混淆矩阵

Fig. 7 Confusion matrix of NASNet-mobile-maize test results

图8 5个玉米品种的特征A：原始图像；B：第1层卷积图像；C：最后1层卷积图像

Fig. 8 Characteristics of 5 corn varietiesA： Original image； B： Image of the first layer of convolution； C： Image of the last layer of convolution

图9 NASNet-mobile-maize测试结果的ROC曲线

Fig. 9 ROC curve of NASNet-mobile-maize test result

参考文献 24

1	闫彬,杨福增,郭文川.基于机器视觉技术检测裂纹玉米种子[J].农机化研究,2020,42(5):181-185, 235.
	YAN B, YANG F Z, GUO W C. Detection of cracked corn seeds based on machine vision technology [J]. Agric. Mech. Res., 2020, 42(5):181-185, 235.
2	吕梦棋,张芮祥,贾浩.基于改进ResNet玉米种子分类方法研究[J].中国农机化学报,2021,42(4):92-98.
	LYU M Q, ZHANG R X, JIA H. Research on classification based on improved ResNet [J]. J. Chin. Agric. Mech., 2021, 42(4):92-98.
3	徐云碧,王冰冰,张健,等.应用分子标记技术改进作物品种保护和监管[J].作物学报,2022,48(8):1853-1870.
	XU Y B, WANG B B, ZHANG J, et al.. Using molecular marker technology to improve the protection and supervision of judgment characteristics [J]. Acta Agron. Sin., 2022, 48(8):1853-1870.
4	冯瑞杰,陈争光,衣淑娟.基于贝叶斯优化的SVM玉米品种鉴别研究[J].光谱学与光谱分析,2022,42(6):1698-1703.
	FENG R J, CHEN Z G, YI S J. Study on SVM standard shujuan based on Bayesian optimization [J]. Spectrosc. Spect. Anal., 2022, 42(6):1698-1703.
5	阳灵燕,张红燕,陈玉峰,等.机器学习在农作物品种识别中的应用研究进展[J].中国农学通报,2020,36(30):158-164.
	YANG L Y, ZHANG H Y, CHEN Y F, et al.. Bulletin on the application of machine learning in agronomic judgment in agronomic judgment [J]. Chin. Agric. Sci. Bull., 2020, 36(30):158-164.
6	李浩光,李卫军,覃鸿,等.基于栈式自编码神经网络的包衣单籽粒玉米品种识别[J].农业机械学报,2017,48():422-428.
	LI H G, LI W J, QIN H, et al.. Characteristic recognition of single-grain coated grains based on stacked Zihong neural network [J]. Trans. Chin. Soc. Agric. Mach., 2017, 48(S1):422-428.
7	马睿,王佳,赵威,等.基于卷积神经网络与迁移学习的玉米籽粒图像分类识别[J].中国粮油学报,2023,38(5):128-134.
	MA R, WANG J, ZHAO W, et al.. Classification and recognition of corn kernel image based on convolutional neural network and transfer learning [J]. J. Chin. Cereals Oils Assoc., 2023, 38(5):128-134.
8	冯晓,张辉,周蕊,等.基于深度学习和籽粒双面特征的玉米品种识别[J].系统仿真学报,2021,33(12):2983-2991.
	FENG X, ZHANG H, ZHOU R, et al.. Characteristics based on deep learning and double-sided features of grains [J]. J. Sys. Sim., 2021, 33(12):2983-2991.
9	刘林.基于深度学习的玉米籽粒品种识别[D].青岛:山东科技大学,2020.
	LIU L. Identification of corn kernel characteristics based on in-depth research [D]. Qingdao: Shandong University of Science and Technology, 2020.
10	TU K, WEN S, CHENG Y, et al.. A non-destructive and highly efficient model for detecting the genuineness of maize variety ‘JINGKE 968’ using machine vision combined with deep learning [J/OL]. Computers Electronics Agric., 2021, 182:106002 [2022-06-12]. .
11	胡艳侠.基于机器视觉的玉米果穗品质检测方法研究[D].西安:长安大学,2017.
	HU Y X.Research on corn ear quality detection method based on machine vision [D]. Xi’an: Chang’an University, 2017.
12	马钦,张佳婧,刘哲,等.基于双路卷积神经网络的玉米制种果穗筛分方法及装置:CN110766053A[P]. 2020-02-07.
13	ZOPH B, VASUDEVAN V, SHLENS J, et al.. Learning transferable architectures for scalable image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition, 2018:8697-8710.
14	黄家才,舒奇,朱晓春,等.基于迁移学习的机器人视觉识别与分拣策略[J].计算机工程与应用,2019,55(8) :232-237.
	HUANG J C, SHU Q, ZHU X C, et al.. Robot vision recognition and sorting strategy based on transfer learning [J].Comput. Eng. Appl., 2019, 55(8):232-237.
15	石祥滨,房雪键,张德园,等.基于深度学习混合模型迁移学习的图像分类[J].系统仿真学报,2016,28(1): 167-173, 182.
	SHI X B, FANG X J, ZHANG D Y, et al.. Image classification based on mixed deep learning model transfer learning [J]. J. Sys. Sim., 2016, 28(1):167-173, 182.
16	王佳,马睿,马德新.基于深度学习的登海605玉米品种真伪鉴别方法研究[J].中国粮油学报,2023, 38(3): 151-157.
	WANG J, MA R, MA D X. Identification method of Denghai 605 maize varieties based on deep learning [J]. J. Chin. Cereals Oils Assoc., 2023, 38(3): 151-157.
17	CHOLLET F. Xception: deep learning with depthwise separable convolutions [C]// Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE, 2017:1800-1807.
18	HE K M, ZHANG X Y, REN S Q, et al.. Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, NV, USA: IEEE, 2016:770-778.
19	SANDLER M, HOWARD A, ZHU M, et al.. MoblieNetV2: inverted residual and linear bottlenecks [C]// Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018:4510-4520.
20	HUANG G, LIU Z, LAURENS V, et al.. Densely connected convolution networks [C]// Proceedings 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, USA: IEEE, 2017:2261-2269.
21	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition [J/OL]. Computer Sci., 2014, arXiv:1409.1556 [2022-06-12]. .
22	杨昊岩,栾涛,韩仲志,等.基于深度学习声谱图分类的“听声识风”[J].华南师范大学学报(自然科学版),2021,53(5):10-16.
	YANG H Y, LUAN T, HAN Z Z, et al.. “Listening to sound and knowing wind” based on deep learning spectrogram classification [J]. J. South China Norm. Univ. (Nat. Sci.), 2021, 53(5):10-16.
23	徐岩,刘林,李中远,等.基于卷积神经网络的玉米品种识别[J].江苏农业学报,2020,36(1):18-23.
	XU Y, LIU L, LI Z Y, et al.. Recognition of corn varieties based on convolutional neural network [J]. Jiangsu J. Agric. Sci., 2020, 36(1):18-23.
24	许景辉,邵明烨,王一琛,等.基于迁移学习的卷积神经网络玉米病害图像识别[J].农业机械学报,2020,51(2):230-236, 253.
	XU J H, SHAO M Y, WANG Y C, et al.. Image recognition of corn diseases based on convolutional neural network based on transfer learning [J]. Trans. Chin. Soc. Agric. Mach., 2020, 51(2):230-236, 253.

[1]	林开颜, 梅飞, 吴军辉, 郭文刚, 陈杰, 司慧萍. 基于计算机视觉的作物病害监测服务平台设计与研究[J]. 中国农业科技导报, 2023, 25(6): 89-96.
[2]	骆丽莎, 廖桂平, 刘凡, 官春云. 基于冠层光谱特征参数的油菜品种识别[J]. 中国农业科技导报, 2021, 23(7): 93-106.
[3]	邵萌1,2,张颖1,2,郭新宇1,2*. 玉米果穗养分运输组织的表型研究进展[J]. 中国农业科技导报, 2016, 18(3): 38-45.
[4]	张大龙,赵武云*,杨婉霞,蒋五洋. 基于组态王的玉米果穗烘干监控系统设计[J]. , 2013, 15(2): 115-119.