基于偏最小二乘法的小麦条锈病潜育期冠层高光谱分析
Canopy hyperspectral features analysis of latent period wheat stripe rust based on discriminant partial least squares

为能够找到更快速、高效的监测早期小麦条锈病的方法，通过获得在0.2、0.1、0.05 mg/mL三种不同浓度小麦条锈菌胁迫下的潜育期小麦冠层光谱数据，利用定性偏最小二乘法建立小麦条锈病潜育期小麦叶片冠层光谱识别模型并分析在3类不同光谱特征（波段、变量、建模比）下的模型准确性及适应性。结果表明，在325~1 075 nm波段内，以伪吸收系数二阶导数[log10（1/R）_2nd.dv]为光谱特征所建模型的平均准确率最高，训练集为97.89%，测试集为92.98%，可优先作为建模时备选的变量参数；在不同波段范围所建模型中，在925~1 075 nm波段内，以伪吸收系数的一阶导数[log10（1/R）]为光谱特征所建模型的平均准确率最高，训练集为98.27%，测试集为94.33%，可优先作为建模时备选的波段范围。表明利用冠层高光谱特征可以实现对小麦条锈病潜育期的定性分析，是一种能早期监测小麦条锈病的损高效方法。
In order to find a better method of monitoring the wheat stripe rust earlier, the canopy hyperspectral data of wheat which was inoculated with three different concentrations(0.2, 0.1, 0.05 mg/mL) in the latent period was collected. With the method of discriminant partial least squares, the adaptability and accuracy of models with different modeling proportions, different spectra features and different wavebands were assessed. The results showed that under the different spectra features and the different modeling proportions conditions, the model with the log10(1/R)’s 2nd derivative’s accuracy was better than others in 325-1 075 nm spectral region(the whole region), the training set’s average accurate rate was 97.28%, and the testing set’s average accurate rate was 92.98%, which could be considered priority as modeling alternative variable parameters. The models with log10(1/R)’s 1st derivative was better than other models in 925-1 075 nm waveband, and the training set’s average accurate rate was 98.27%, and the testing set’s average accurate rate was 94.33%, which could be considered priority as modeling waveband. The results indicated that the qualitative identification of wheat stripe rust in the latent period could be implemented based on hyperspectral data at the canopy level, and it was simple, rapid, nondestructive and high-efficiency, so it was provided a new method for early monitoring wheat stripe rust.