基于地面高光谱遥感的降香黄檀黑痣病病情指数反演

利用美国Spectra Vista Corporation(以下均用简称SVC)HR-1024i非成像高光谱仪采集不同病情程度的降香黄檀冠层光谱数据,并结合地面同步调查获得的降香黄檀黑痣病病情指数数据,对光谱数据进行重叠校正(scan matching/overlap correction)和白光板反射率校正(white plate reflectance correction)。采用主成分分析法(PCA法)对与降香黄檀黑痣病病情指数相关性较高的敏感波段进行降维。利用53个训练集,将敏感波段和PCA法处理后的敏感波段分别作为输入变量,训练降香黄檀黑痣病的BP神经网络。两种输入变量建立的神经网络计算出的预测值与实际值之间的决定系数(R2)均达到99%。利用27个验证集做进一步精度检验,结果表明,通过这两种输入变量训练的BP神经网络,得到的预测值与实际值之间的决定系数(R2)分别为0.951 9和0.706 0,均方根误差(RMSE)分别为5.998 0和12.919 3。直接以敏感波段作为变量输入和PCA法处理后的敏感波段作为变量输入训练BP神经网络是一种有效的方法,其中,直接以敏感波段作为变量输入精度更高。

Dalbergia odorifera black scurf disease index inversion based on ground hyperspectral technology

Canopy spectral data of Dalbergia odorifera were collected according to different disease incidences, using Spectra Vista Corporation (SVC) HR-1024i un-imaging hyperspectral of America, then scan matching/overlap correction and white plate reflectance correction of spectral data were completed based on the disease index of D. odorifera black scurf obtained simultaneously in the field. Principal component analysis (PCA) was applied to conduct dimension-reduction of sensitive wave band which highly related to disease index. Both sensitive wave bands from 53 training sets before and after processing by PCA were chosen as input variables for training BP neural network of D. odorifera black scurf. The results showed that both coefficients of determination (R2) between the predictive values from BP neural network established by above two variables and the actual values were to 99%. Further accuracy test by using 27 validation sets showed that the coefficients of determination (R2) between the predictive value and the actual value were up to 0.951 9 and 0.706 0, and the root mean square errors (RMSE) were 5.998 0 and 12.919 3. The results indicated that both methods of training BP neural network by using sensitive wave bands directly and after treatment by PCA as variables were all effective ways, of which using sensitive wave bands directly was more accurate