欢迎访问《麦类作物学报》编辑部网站

作者	单位
张宏，李卫国，张晓东，李伟，马廷淮，韩振强	1.江苏大学农业工程学院，江苏镇江 212013；2.江苏省农业科学院农业信息研究所，江苏南京 210014；3.江苏大学流体机械工程技术研究中心，江苏镇江 212013；4.南京信息工程大学，江苏南京 210044

摘要点击次数: 235

全文下载次数: 178

中文摘要:

为了快速、准确地估测大田冬小麦茎蘖数（stem & tiller number，STN），在江苏省盐城市大丰区、泰州泰兴市和宿迁市沭阳县布设冬小麦STN遥感估测试验，获取了冬小麦拔节期冠层红光波段反射率（red band reflectance，β_red）、近红外波段反射率（near infrared band reflectance，β_nir）、比值植被指数（ratio vegetation index，RVI）、归一化差值植被指数（normalized difference vegetation index，NDVI）、差值植被指数（differential vegetation index，DVI）、阴影植被指数（shadow vegetation index，SVI）和STN数据，通过分析多个遥感光谱指标（β_red、β_nir、RVI、NDVI、DVI、SVI）与STN之间的相关性，优选冬小麦STN的敏感光谱指标，再基于敏感光谱指标分别建立冬小麦STN的BP神经网络估测模型（STN_BP估测模型）和多元线性回归估测模型（STN_MLR估测模型），并对模型预测精度进行验证。结果表明，β_red、β_nir、RVI、NDVI、DVI和SVI与冬小麦STN之间均存在不同程度的相关性，其相关系数依次表现为β_red（0.337）<β_nir（0.375）BP估测模型和STN_MLR估测模型的输入变量。模型精度验证显示，STN_BP估测模型的决定系数（coefficient of determination，R²）为0.758，均方根误差（root mean square error，RMSE）为2.169×10⁶个·hm^-2，平均相对误差（average relative error，ARE）为13.7%；STN_MLR估测模型的R²为0.599，RMSE为3.110×10⁶个·hm^-2，ARE为20.0%。STN_BP估测模型的估测精度优于STN_MLR估测模型，说明利用多遥感光谱敏感特征指标和BP神经网络建立的冬小麦STN_BP估测模型能够有效满足大田冬小麦茎蘖数的估测要求。

英文摘要:

In order to quickly and accurately estimate stem and tiller number (STN) of winter wheat in field, the remote sensing estimation experiment of winter wheat stem and tiller number was carried out in Dafeng District of Yancheng City, Taixing City of Taizhou City and Shuyang County of Suqian City, Jiangsu Province. The red band reflectance (β_red) and near infrared band reflectance (β_nir) of winter wheat canopy, ratio vegetation index (RVI), normalized difference vegetation index (NDVI), differential vegetation index (DVI), shadow vegetation index (SVI), and winter wheat STN data at jointing stage were obtained. By analyzing the correlation between several remote sensing spectral indices (β_red, β_nir, RVI, NDVI, DVI, SVI) and winter wheat STN, the sensitive spectral indices of winter wheat STN were optimized. The BP neural network estimation model (STN_BP estimation model) and the multiple linear regression estimation model (STN_MLR estimation model) of winter wheat STN were established based on sensitive spectral indices, and the prediction accuracy of the models was verified. The results showed that β_red, β_nir, RVI, NDVI, DVI, and SVI had different degrees of correlation with winter wheat STN. The correlation coefficients ranked as β_red (0.337) <β_nir (0.375) < DVI (0.423) < RVI (0.446) < SVI (0.447) < NDVI (0.470), among which the RVI, NDVI, DVI and SVI were selected as the input variables to establish the STN_BP and STN_MLR estimation models. The model accuracy verification showed that the coefficient of determination (R²), the root mean square error (RMSE), and the average relative error (ARE) of STN_BP estimation model were 0.758, 2.169×10⁶ pieces·hm^-2, and 13.7%, respectively. The R², RMSE, and ARE of STN_MLR estimation model were 0.599, 3.110×10⁶ pieces·hm^-2 and 20.0%, respectively. By comparison, the estimation effect of STN_BP estimation model is better than that of STN_MLR estimation model, indicating that the estimation model of winter wheat STN_BP estimation model based on multiple remote sensing spectral sensitive feature index and BP neural network can effectively meet the estimation requirements of STN of winter wheat in field.

查看全文查看/发表评论下载PDF阅读器

关闭