用高光谱成像技术检测柑橘红蜘蛛为害叶片的色素含量

Pigment content measurement for citrus red mite infected leaf using hyper-spectral imaging technology

为解决传统理化法检测柑橘树叶片受红蜘蛛为害后色素含量变化时存在的工作量大、效率低等问题,该文研究应用高光谱成像技术检测柑橘红蜘蛛为害叶片色素含量的方法。研究中对比了正常叶片与受害叶片的原始光谱以及原始光谱一阶微分曲线的差异,寻找反映叶片色素含量变化的特征波段;分析了特征波段反射率比值与叶片色素间相关性;采用单变量线性回归法分析了常用植被指数预测叶片色素含量的效果;采用逐步回归分析法建立了叶片色素含量预测模型,并对模型预测效果进行了F检验。结果表明:常用植被指数预测叶片色素含量结果不理想;选取的667/522、667/647和522/647 nm等3个特征波段反射率比值与叶片3种色素含量间具有较高的相关性;用于建立叶片色素含量预测模型的最佳特征波段反射率比值为667/522和667/647 nm,所建立的模型可较好地预测健康及受害叶片的叶绿素a、叶绿素b和类胡萝卜素含量。

In order to solve the high workload and low efficiency problems while measuring the pigment content variation of citrus red mite infested leaves using the traditional physical and chemical methods, a novel pigment content measurement method for citrus red mite infested leaf using the hyper-spectral imaging technology was studied in this paper. In the research, 400 healthy leaves and 400 sick leaves were included as the test samples in which 350 healthy leaves and 350 sick leaves were utilized for model establishment and the other 50 leaves of each type were used for a model test. Each leaf’s original spectrum and its first order deviation in its particular healthy and sick area were acquired to investigate the characteristic spectrum bands which could mostly reflect the variation of leaf pigment content. The correlation between characteristic spectrum band ratios and pigment content was analyzed. An univariate linear regression method was applied to analyze the pigment content prediction effect using the common vegetation indexes. A leaf pigment content prediction model was established, using the stepwise regression method, and the model’s prediction ability was tested using the F test. Experimental results indicated that it is not satisfactory using the common vegetation indexes to predict leaf pigment content since they are not specially selected for citrus trees. The selected three characteristic spectrum band ratios of 667/522, 667/647, and 522/647 nm, each of which has a high correlation with a leaf’s three types of pigment content, were applied in the stepwise regression method to establish pigment content prediction models. Two out of three of the characteristic spectrum band ratios of 667/522 and 667/647 nm, which gave the best performance, were used as independent values for model establishment. The F test results indicated that the established models could preferably predict both healthy and sick leaves chlorophyll a, chlorophyll b, and carotenoid content. The selected characteristic bands, as well as the established prediction models, could be used as the foundation to further study the citrus red mite infestation fast detection methods and techniques.