基于机器视觉的茄子种子发芽率预测方法  

Prediction method of eggplant seed germination rate based on machine vision

作　　者：海妍 张君 张东方 李玉超 刘景艳[1] 范晓飞 索雪松[1] HAI Yan;ZHANG Jun;ZHANG Dongfang;LI Yuchao;LIU Jingyan;FAN Xiaofei;SUO Xuesong(College of Mechanical and Electrical Engineering,Hebei Agricultural University,Baoding 071001,China;College of Horticulture,Hebei Agricultural University,Baoding 071001,China)

机构地区：[1]河北农业大学机电工程学院,河北保定071001 [2]河北农业大学园艺学院,河北保定071001

出　　处：《河北农业大学学报》2023年第6期103-108,共6页Journal of Hebei Agricultural University

基　　金：国家自然科学基金面上项目(32070572);河北省高层次人才资助项目(E2019100006);河北省重点研发计划项目(20327403D);河北农业大学引进人才项目(YJ201847)。

摘　　要：为了研究茄子种子的发芽能力,提出了1种基于机器视觉的快速无损预测茄子种子发芽率的研究方法。为了探索种子发芽能力与表型特征之间的关系,本研究选用具有五通道的多光谱相机对新鲜的茄子种子进行图像信息采集。本试验设计了2种机器学习模型,含基于所提取表型数据的预测模型与基于五通道多光谱图像的深度学习预测模型,对茄子种子的发芽率进行预测。首先利用颜色空间转换、形态学处理等图像处理算法对种子表型特征数据进行提取,结合主成分分析以及连续投影算法对数据进行降维分析,建立了基于支持向量机的种子发芽率预测模型。同时建立了基于原始多光谱图像的卷积神经网络预测模型。通过2种模型的对比,基于多光谱成像技术的深度学习模型在对茄子种子的发芽率预测有着更高的准确率,其验证集准确率为84.3%。本研究对茄子种子图像代表性特征的选择和识别样本的简化使得分类系统更符合实际生产需要,在茄子种子发芽率的预测中有着积极的意义。In order to study the germination ability of eggplant seeds,this paper proposed a research method based on machine vision to quickly and non-destructively predict the germination rate of eggplant seeds.In order to explore the relationship between seed germination ability and phenotypic characteristics,this study selected a five-channel multispectral camera to collect image information of fresh eggplant seeds.In this experiment,two machine learning models were designed,including a prediction model based on the extracted phenotypic data and a deep learning prediction model based on five-channel multispectral images to predict the germination rate of eggplant seeds.Firstly,image processing algorithms including color space conversion and morphological processing were used to extract the seed phenotypic characteristic data,and the principal component analysis and continuous projection algorithm were combined to perform dimensionality reduction analysis on the data,and a seed germination rate prediction model based on support vector machine was then established.At the same time,a convolutional neural network prediction model based on the original multispectral image was established.Through the comparison of the two models,the deep learning model based on multispectral imaging technology displayed a higher accuracy in predicting the germination rate of eggplant seeds,and its verification set accuracy rate was 84.3%.In this study,the selection of representative features of eggplant seed images and the simplification of identification samples made the classification system more in line with actual production needs,which had positive significance in the prediction of eggplant seed germination rate.

关 键 词：多光谱成像 机器学习 茄子种子 发芽率 

分 类 号：S24[农业科学—农业电气化与自动化]