| 摘要: |
| 针对大豆联合收割机械作业含杂率在线检测手段缺乏的问题，以亚丰4YZL-5S联合收获机机械化收获的大豆样本为研究对象，在室内测定大豆样本的含杂率；利用ASD FieldSpec 4 Wide-Res型地物光谱仪测量大豆样本的光谱数据，经数据预处理和数学变换后获得2种光谱指标，即原始光谱数据（REF）和原始光谱经倒数之对数预处理后的数据（LR），应用波段间自相关分析筛选出不同指标的大豆样本光谱的特征波长，并采用支持向量机回归分析构建基于不同指标的大豆样本含杂率的反演模型，在此基础上对反演结果进行精度验证和比较。试验结果表明：各预处理条件下的大豆含杂率敏感波段不同，其中REF的特征波段为512，738，851，1 104，2 003，2 179 nm；LR的特征波段为519，637，820，924，1 121，1 933，2 050，2 138 nm。本研究建立的含杂率反演模型的建模决定系数>0.86，验证决定系数>0.79，均方根误差<0.32，相对分析误差>1.7，表明模型具有较强的拟合效果和预测能力。相比较而言，利用REF建立的反演模型的反演效果略优于LR。本研究建立的大豆样本含杂率光谱反演模型能够实现含杂率的在线预测，为大豆机械化作业中含杂率的在线快速监测提供了新途径。 |
| 关键词: 大豆 含杂率 机械化作业 支持向量机回归分析 高光谱 |
| DOI：10.11841/j.issn.1007-4333.2019.09.17 |
| 投稿时间：2019-01-16 |
| 基金项目:国家重点研发计划重点专项（2017YFD0700305）；中央级公益性科研院所基本科研业务费专项（S201818）；工信部2017年智能制造新模式应用项目资助（20170829R2） |
|
| Inversion model of soybean impurity rate based on hyperspectral |
|
CHEN Man, XU Jinshan, JIN Chengqian, ZHANG Guanyue, NI Youliang
|
| (Nanjing Research Institute for Agricultural Mechanization, Ministry of Agriculture and Rural Affairs, Nanjing 210014, China) |
| Abstract: |
| In view of the lack of online detection means for the impurity rate of soybean combined harvesting machinery operation,the soybean samples harvested by Yafeng 4YZL-5S combine harvester were taken as the research object,and the online recognition arithmetic of impurity rate based on spectral inversion was carried out in this study.Firstly,the impurity rate of soybean samples was determined in laboratory.Second,the ASD FieldSpec 4 wide-res was used to measure the spectral data of soybean samples.After data preprocessing and mathematical transformation,two spectral indexes were obtained,the original spectral data (REF) and the data after reciprocal and logarithmic preprocessing of the original spectrum (LR).Thirdly,the characteristic wavelength of the spectrum of soybean samples with different indexes was selected by using the inter-band autocorrelation analysis.Finally,support vector machine (SVM) regression analysis was used to construct the inversion model of soybean sample impurity rate based on different indexes.On this basis,the precision of the inversion results was verified and compared.The results showed that the sensitive bands of soybean impurity content were different under different pretreatment conditions.The characteristic bands of REF were 512,738,851,1 104,2 003,and 2 179 nm,respectively.The characteristic bands of LR are respectively 519,637,820,924,1 121,1 933,2 050,and 2 138 nm.The modeling determination coefficient of the inversion model established in this study is greater than 0.86,the verification determination coefficient is greater than 0.79,the root-mean-square error is less than 0.32,and the relative analysis error is greater than 1.7,indicating that the model has a strong fitting effect and prediction ability.In comparison,the inversion effect of the inversion model established by REF is slightly better than that of LR.The spectral inversion model established in this study can realize online prediction of the impurity rate of soybean samples,providing a new way for online rapid monitoring of the impurity rate in soybean mechanized operation. |
| Key words: soybean impurity rate mechanized operations support vector regression hyperspectral remote sensing |
