基于卷积神经网络的列当种子发芽识别方法

doi:10.11686/cyxb2023358

摘要/Abstract

摘要：

列当是一种极具破坏力的寄生性杂草，目前难以用常规方法根除。通过使用发芽刺激物可以诱导列当“自杀性萌发”，因此识别特定发芽刺激物下的种子发芽率至关重要。针对显微镜下人眼视觉对列当种子发芽判断标准不统一和耗时耗力等问题，本研究提出一种基于卷积神经网络的列当种子发芽识别算法。首先，培育列当种子并采用显微镜采集发芽和未发芽的种子构建列当图像库；然后，通过搭建卷积神经网络来提取列当图像的特征实现种子发芽识别，通过试验对比优化超参数，得到适合列当种子识别的OB-Net模型；最后，通过试验图像验证，本研究模型可达到95.2%的识别准确率；与现有主流网络模型进行对比证明了构建的OB-Net模型在列当种子识别方面具有最高的精度和较快检测速度。本研究的列当种子发芽识别方法可以为相关发芽刺激物的研究提供有效的理论支撑。

关键词: 列当种子, 发芽识别, 卷积神经网络, 特征提取

Abstract:

Broomrape （Orobanche spp.） is an exceedingly pernicious parasitic weed that is difficult to eradicate using conventional methods. Inducing “suicidal germination” in broomrape seeds through the application of germination stimulants is a crucial control method. However， the current method to evaluate broomrape seed germination based on human visual inspection using a microscope is time-consuming and produces inconsistent results. To address these issues， we propose a broomrape seed germination recognition algorithm based on convolutional neural networks. First， we cultivated broomrape seeds and collected images of germinated and ungerminated seeds under a microscope to construct a broomrape image library. Then， we developed a convolutional neural network， named OB-Net， to extract features from broomrape images and recognize seed germination. Through comparative analysis and optimization， we carefully selected the hyperparameters of the OB-Net model. Our experimental results demonstrated that the model achieved a recognition accuracy of 95.2%. Comparative analysis with existing mainstream network models confirmed that the proposed OB-Net model exhibited the highest accuracy and fastest detection speed in recognizing germinated broomrape seeds. The broomrape seed germination recognition method proposed in this study offers effective theoretical support for further research on other seeds and germination stimulants.

Key words: broomrape seeds, germination recognition, convolutional neural networks, feature extraction

沈祺嵘, 严俊, 叶晓馨, 桑玉莹, 单启玲, 张琪藤. 基于卷积神经网络的列当种子发芽识别方法[J]. 草业学报, 2024, 33(8): 112-121.

Qi-rong SHEN, Jun YAN, Xiao-xin YE, Yu-ying SANG, Qi-ling SHAN, Qi-teng ZHANG. A broomrape seed germination recognition method based on convolutional neural networks[J]. Acta Prataculturae Sinica, 2024, 33(8): 112-121.

图/表 8

图1 发芽与未发芽的列当种子红色不规则框选部分为胚根The red irregularly boxed part is the radical.

Fig.1 Germinated and ungerminated Orobanche seed

图2 用于特征提取和识别的OB-Net的整体结构图中数字以像素为单位，ReLU 为网络所使用的激活函数Numbers in the figure are in pixels， ReLU （rectified linear unit） is the activation function used by the network.

Fig.2 The overall structure of OB-Net for feature extraction and recognition

图3 学习率对模型训练效果的影响

Fig.3 The effect of learning rate on model training results

图4 卷积层对模型训练效果的影响

Fig.4 The effect of convolutional layers on model training results

图5 图像尺寸对模型训练效果的影响

Fig.5 The effect of image size on model training results

图6 数据增广对模型训练效果的影响

Fig.6 Effects of data augmentation on training accuracy and training loss

表1 基于特征图和热图的列当种子预测

Table 1 Orobanche seed prediction results based on feature maps and heat maps

项目 Item	预测标签 Prediction label	手动标签 Manual labels
a	未发芽Ungermination	未发芽Ungermination
b	未发芽Ungermination	未发芽Ungermination
c	未发芽Ungermination	未发芽Ungermination
d	未发芽Ungermination	未发芽Ungermination
e	发芽Germination	发芽Germination
f	发芽Germination	发芽Germination
g	发芽Germination	发芽Germination
h	发芽Germination	发芽Germination
i	未发芽Ungermination	发芽Germination
j	发芽Germination	未发芽Ungermination
混淆矩阵Confusion matrix	$0.954 0.046 0.05 ??? 0.95 ???$
验证准确性Verify accuracy （%）	95.2

表1 基于特征图和热图的列当种子预测

Table 1 Orobanche seed prediction results based on feature maps and heat maps

项目 Item	预测标签 Prediction label	手动标签 Manual labels
a	未发芽Ungermination	未发芽Ungermination
b	未发芽Ungermination	未发芽Ungermination
c	未发芽Ungermination	未发芽Ungermination
d	未发芽Ungermination	未发芽Ungermination
e	发芽Germination	发芽Germination
f	发芽Germination	发芽Germination
g	发芽Germination	发芽Germination
h	发芽Germination	发芽Germination
i	未发芽Ungermination	发芽Germination
j	发芽Germination	未发芽Ungermination
混淆矩阵Confusion matrix	$0.954 0.046 0.05 ??? 0.95 ???$
验证准确性Verify accuracy （%）	95.2

表2 不同卷积神经网络模型性能对比

Table 2 Performance comparison of different convolutional neural network models

模型 Model	训练准确率Train accuracy （%）	参数量 Parameter quantity （No.）	单图预测时间Single graph prediction time （ms）	测试准确率 Test accuracy （%）
LeNet	91.1	1.05×10⁵	1.21	90.4
VGG11	92.6	3.50×10⁷	6.50	91.9
AlexNet	94.6	5.80×10⁶	6.20	95.1
ResNet18	95.7	1.17×10⁷	8.43	94.7
OB-Net	95.3	1.21×10⁶	2.50	95.2

参考文献 27

1	Guo S C， Miao H M， Li S P， et al. Research advances in the breeding study of sunflower resistance to Orobanche. Crops， 2022， 38（3）： 27-32.
	郭树春，苗红梅，李素萍，等. 向日葵抗列当育种研究进展. 作物杂志， 2022， 38（3）： 27-32.
2	Ye X X， Zhang M， Dong S， et al. Conditioning duration and agents involved in broomrape seeds responding to germination stimulants. Plant Growth Regulation， 2017， 81（2）： 221-230.
3	Fernández-Aparicio M， Cimmino A， Soriano G， et al. Assessment of weed root extracts for allelopathic activity against Orobanche and Phelipanche species. Phytopathologia Mediterranea， 2021， 60（3）： 455-466.
4	Chen X F. Causes and comprehensive prevention and control measures of sunflower in Dingbian County. Modern Agricultural Sciences and Technology， 2010， 533（15）： 215， 217.
	陈秀芳. 定边县向日葵列当发生原因及综合防治措施. 现代农业科技， 2010， 533（15）： 215， 217.
5	Ma Y Q， Dong S Q， Ren X X， et al. Parasitic weed Orobanche spp. and perspective of its control measures. Chinese Journal of Biological Control， 2012， 28（1）： 133-138.
	马永清，董淑琦，任祥祥，等. 列当杂草及其防除措施展望. 中国生物防治学报， 2012， 28（1）： 133-138.
6	Cui H X， Wang N， Long X Q， et al. Review of the species， hazard and management status of Orobanche L. in Xinjiang. Plant Quarantine， 2020， 34（3）： 20-24.
	崔华星，王宁，龙宣杞，等. 新疆列当种类、危害与防治现状. 植物检疫， 2020， 34（3）： 20-24.
7	Fernández-Aparicio M， Reboud X， Gibot-Leclerc S. Broomrape weeds//Underground mechanisms of parasitism and associated strategies for their control： a review. Frontiers in Plant Science， 2016， 7： 135.
8	Song W J， Zhou W J， Jin Z L， et al. Germination response of Orobanche seeds subjected to conditioning temperature， water potential and growth regulator treatments. Weed Research， 2005， 45（6）： 467-476.
9	Zhang S， Zhang S， Zhang C， et al. Cucumber leaf disease identification with global pooling dilated convolutional neural network. Computers and Electronics in Agriculture， 2019， 162： 422-430.
10	Majumdar S， Jayas D S. Classification of cereal grains using machine vision： I. Morphology models. Transactions of the ASAE， 2000， 43（6）： 1669-1675.
11	He X， Wang X M， Zhang H， et al. Convolutional neural network-based image recognition of forage seeds. Pratacultural Science， 2022， 39（11）： 2338-2349.
	何欣，王雪萌，张涵，等. 基于卷积神经网络的牧草种子图像识别. 草业科学， 2022， 39（11）： 2338-2349.
12	Özkan K， Işık Ş， Yavuz B T. Identification of wheat kernels by fusion of RGB， SWIR， and VNIR samples. Journal of the Science of Food and Agriculture， 2019， 99（11）： 4977-4984.
13	Zhu S S， Zhou L， Gao P， et al. Near-infrared hyperspectral imaging combined with deep learning to identify cotton seed varieties. Molecules， 2019， 24（18）： 3268.
14	Wan L Y. Research on automatic crop seed counting method based on image recognition. Chengdu： Chengdu University， 2020.
	万路瑶. 基于图像识别的作物种子自动计数方法研究. 成都：成都大学， 2020.
15	Khan M A， Akram T， Sharif M， et al. An automated system for cucumber leaf diseased spot detection and classification using improved saliency method and deep features selection. Multimedia Tools and Applications， 2020， 79（25）： 18627-18656.
16	Lin C C， Shan C， Zhao G S， et al. Review of image data augmentation in computer vision. Journal of Frontiers of Computer Science and Technology， 2021， 15（4）： 583-611.
	林成创，单纯，赵淦森，等. 机器视觉应用中的图像数据增广综述. 计算机科学与探索， 2021， 15（4）： 583-811.
17	Moral J， Lozano-Baena M D， Rubiales D. Temperature and water stress during conditioning and incubation phase affecting Orobanche crenata seed germination and radicle growth. Frontiers in Plant Science， 2015， 6： 408.
18	Wen X. Modeling and performance evaluation of wind turbine based on ant colony optimization-extreme learning machine. Applied Soft Computing， 2020， 94： 106476.
19	Yang H Y， Huang Q， Sun A Z， et al. Rice seed image classification and recognition based on CNN_SVM. Journal of the Chinese Cereals and Oils Association， 2021， 36（12）： 144-150.
	杨红云，黄琼，孙爱珍，等. 基于卷积神经网络和支持向量机的水稻种子图像分类识别. 中国粮油学报， 2021， 36（12）： 144-150.
20	Hou J M， Yao E C， Zhu H J. Classification of castor seed damage based on convolutional neural network. Transactions of the Chinese Society for Agricultural Machinery， 2020， 51（S1）： 440-449.
	侯俊铭，姚恩超，朱红杰. 基于卷积神经网络的蓖麻种子损伤分类研究. 农业机械学报， 2020， 51（S1）： 440-449.
21	Long M S， Ouyang C J， Liu H， et al. Image recognition of Camellia oleifera disease based on convolutional neural network & transfer learning. Transactions of the Chinese Society of Agricultural Engineering， 2018， 34（18）： 194-201.
	龙满生，欧阳春娟，刘欢，等. 基于卷积神经网络与迁移学习的油茶病害图像识别. 农业工程学报， 2018， 34（18）： 194-201.
22	Li Y D， Hao Z B， Lei H. Survey of convolutional neural network. Journal of Computer Applications， 2016， 36（9）： 2508-2515.
	李彦冬，郝宗波，雷航. 卷积神经网络研究综述. 计算机应用， 2016， 36（9）： 2508-2515.
23	Zhang K， Xu Z， Dong S， et al. Identification of peach leaf disease infected by Xanthomonas campestris with deep learning. Engineering in Agriculture， Environment and Food， 2019， 12（4）： 388-396.
24	Srivastava N， Hinton G， Krizhevsky A， et al. Dropout： a simple way to prevent neural networks from overfitting. The Journal of Machine Learning Research， 2014， 15（1）： 1929-1958.
25	Lecun Y， Bottou L， Bengio Y， et al. Gradient-based learning applied to document recognition. Proceedings of the IEEE， 1998， 86（11）： 2278-2324.
26	Selvaraju R R， Cogswell M， Das A， et al. Grad-CAM： Visual explanations from deep networks via gradient-based localization. International Journal of Computer Vision， 2020， 128（2）： 336-359.
27	Wang H， Zhao W T， Chen L F， et al. Collection and preprocessing methods of broomrape （Orobanche spp. and Phelipanche spp.） seeds. Journal of Weed Science， 2016， 34（1）： 22-25.
	王焕，赵文团，陈连芳，等. 列当（Orobanche spp. and Phelipanche spp.）种子的采集与预处理方法. 杂草学报， 2016， 34（1）： 22-25.