基于近红外光谱技术及ELM...不同生长阶段米象的分类识别_鲁玉杰.pdf

2023 年 2 月第 44 卷第 1 期河南工业大学学报(自然科学版)Journal of Henan University of Technology(Natural Science Edition)Feb.2023Vol.44 No.1收稿日期:2022-03-10基金项目:国家重点研发计划项目(2019YFC1605304)作者简介:鲁玉杰(1971),女,河南南阳人,教授,研究方向为储粮害虫生态学和分子生态学,E-mail:。基于近红外光谱技术及 ELM 对小麦中不同生长阶段米象的分类识别鲁玉杰1,2,王文敬1,张俊东1,王争艳1,卢少华11.河南工业大学 粮油食品学院,河南 郑州 4500012.江苏科技大学 粮食学院,江苏 镇江 212100摘要:对粮食中隐蔽性害虫的早期诊断和检测,不仅可以减少因害虫取食造成的粮食产后损失,还可以减少化学药剂的使用,对于保证粮食品质和减少环境污染具有重要的意义。基于近红外光谱技术与极限学习机(ELM)构建小麦中不同生长阶段米象的分类识别模型,采集未感染小麦和感染米象小麦的近红外光谱数据,选择 SNV+De-trending 的组合对原始光谱数据进行预处理,使用主成分分析(PCA)方法对光谱数据进行降维特征提取,利用 ELM 和支持向量机(SVM)建立分类识别模型。结果表明:ELM模型训练时间仅需 0.062 5 s,总体分类准确率为 90%,0、6、24 和 27 d 的识别率为 100%,1020 d 的幼虫期识别率偏低,20 d 时识别率最低,为 65%;SVM 模型运行时间为 3.38 s,分类准确率为 85.42%,ELM 模型较 SVM 模型的运行时间和分类准确率都有所提高。因此,ELM 分类识别模型能够快速准确地判断小麦有无米象,以及分类识别小麦中不同生长发育阶段的米象。关键词:近红外;隐蔽性害虫;极限学习机;分类;米象;早期诊断中图分类号:TS210文献标志码:A文章编号:1673-2383(2023)01-0104-08DOI:10.16433/j.1673-2383.2023.01.014Classification and recognition of Sitophilus oryzae in different growth stages of wheat based on near-infrared spectroscopy and ELMLU Yujie1,2,WANG Wenjing1,ZHANG Jundong1,WANG Zhengyan1,LU Shaohua11.College of Food Science and Engineering,Henan University of Technology,Zhengzhou 450001,China2.School of Grain Science and Technology,Jiangsu University of Science and Technology,Zhenjiang 212100,ChinaAbstract:Early diagnosis and detection of hidden pests in grain could not only reduce the post-production losses of grain caused by pest feeding,but also reduce the use of chemicals,which are important for main-taining grain quality and reducing environmental pollution.In this paper,a classification and identification model of Sitophilus oryzae(S.oryzae)in wheat at different growth stages was constructed based on near-infrared spectroscopy and extreme learning machine(ELM),and the near-infrared spectral data of uninfect-ed wheat and infected S.oryzae wheat were collected.The images of S.oryzae in different growth and development stages were obtained by X-ray imaging technology,and the development period of S.oryzae was obtained through the images(egg stage at 0-9 d,larval stage at 10-20 d,pupa stage at 21-26 d,and adult stage at 27-30 d),the wheat with full grains was selected and collected by near infrared spectroscopy to obtain the spectral data of uninfected samples,and then wheat was infested with S.oryzae adults.After 48 hours,the S.oryzae adults were taken out,and the samples were collected by near-infrared spectrum on 第 44 卷第 1 期鲁玉杰,等:基于近红外光谱技术及 ELM 对小麦中不同生长阶段米象的分类识别6,10,14,17,20,24,and 27 day of the experiment to obtain uninfected wheat and near-infrared spectral data of wheat infected with S.oryzae wheat.When modeling using the original spectral data,the classifica-tion accuracy of the ELM model was 78.75%.After preprocessing,the classification accuracy of the ELM model reached 85%,and then the principal component analysis(PCA)method was used to perform di-mension reduction feature extraction on the spectral data.When the target dimension was 120 dimensions,the accuracy of the ELM classification and recognition model was 90%,the classification recognition rate increased by 12.5%.The experimental results showed that the appropriate preprocessing method and PCA dimensionality reduction feature extraction could effectively improve the classification accuracy of ELM model,and the training time was only 0.062 5 s,the overall classification accuracy reached 90%,the rec-ognition rates were 100%on 0,6,24,and 27 days,and lower on 10-20 days of larval stage,and the rec-ognition rate was the lowest at 20 days,which was 65%.Compared with the performance of ELM and SVM in this experiment,the training time of SVM model was 3.38 s,the overall classification accuracy reached 85.42%,on 0,10,17,and 24 days,the recognition rate was 90%,at 6 days,the recognition rate was 80%,at 14 days,the recognition rate was 55%,and at 20 and 27 days,the recognition rate was 85%.The results showed that the classification effect of ELM model was better than that of SVM model.There-fore,ELM classification and recognition model could quickly and accurately determine insect-free and insect-containing wheat and classify the S.oryzae at different growth and development stages.The classifi-cation and identification of S.oryzae has potential practical value for early detection of hidden pests in grain.The data of this study comes from laboratory conditions.In the future,more data can be collected from actual production to strengthen the classification model and increase the accuracy of the model.The optimized ELM was used to further improve the classification and recognition efficiency and accuracy of the model.On the basis of this paper,a classification and identification model of a variety of hidden pests was established to provide a reference for the intelligent detection of pests in the construction of intelligent grain depots.Key words:NIRS;hidden pests;ELM;classification;Sitophilus oryzae;early diagnosis我国是世界上最大的粮食生产国,同时也是最大的粮食储藏国,每年的粮食储存量高达年产量的一半以上1-2,因此,做好粮食的安全储备工作