论文部分内容阅读
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。
Spectral data of D. officinale from different conditions were collected by non-imaging HR-1024i non-imaging hyperspectral spectrometer from Spectra Vista Corporation (hereinafter referred to as SVC), and combined with the disease index data of Dalbergia striatellus Spectral data were subjected to scan matching / overlap correction and white plate reflectance correction. Principal component analysis (PCA) was used to reduce the sensitive bands that had a higher correlation with disease index of Dalbergia davidiana. Fifty-three training sets were used to train the BP neural network of Dalbergia henryi disease, using sensitive bands and sensitive bands treated by PCA as input variables respectively. The determination coefficient (R2) between the predicted value and the actual value calculated by the neural network established by the two input variables reaches 99%. 27 verification sets are used for further accuracy test. The results show that the decision coefficients (R2) between the predicted value and the actual value obtained by these two input variables are 0.951 9 and 0.706 0, respectively. The mean square The root-mean-square error (RMSE) was 5.998 0 and 12.919 3, respectively. It is an effective way to directly input the sensitive band as a variable input and the sensitive band after PCA processing as a variable to input training BP neural network, in which the sensitive band is directly input as the variable with higher accuracy.