棉花枯萎病的光谱识别

Using Spectral Data to Distinguish Cotton Fusarium Wilt Disease

采用"积分球—光谱仪"联用技术测量了健康棉叶和感染了枯萎病的棉叶的光谱反射率,发现健康棉叶与感病棉叶在光谱曲线上有很高的可区分性。用健康棉叶的波谱带作为分类器分别对"健康—发病期"和"健康—潜病期"两组棉叶的波谱集合进行分类,总体分类精度分别为100%和92%。将测得的光谱数据转化为LandsatTM卫星多光谱数据,同样用健康棉叶的光谱带对以上两组波谱集进行分类,总体分类精度分别为96%和92%。试验结果为遥感技术在监测棉花枯萎病上的应用提供了理论支持。

Fusarium wilt of cotton is a widespread and destructive disease which is caused by the fungus pathogen Fusarium oxysporium f. sp. vasindectum. As a new technique, remote sensing can play a great role in precision agriculture, such as crop production evaluation, crop growth monitor, and so on. The purpose of this study is to evaluate the probability of using remote sensing technique to distinguish healthy cotton and cotton infected by Fusarium wilt disease based on the spectrum theory. The study was executed in the experimental field of the Institute of Plant Protection, Chinese Academy of Agricultural Sciences on 10th June,2005, when the cotton plants had about 5 to 7 leaves. The cotton variety were EJing 1, which were grouped in healthy and infected. The second and the third new fully opened leaves on the top of 13 healthy cotton plants and 24 infected cotton plants were selected to measure their spectrum reflectivity, respectively. For infected plants, their leaves were classified to damaged and not damaged categories according to their appearance. This study used spectro- radiometer-integrating sphere to acquire reflective spectrum data. Using this technique, the shortcomings caused by natural light and timing can be avoided. The integrating sphere used in this study is LI- COR 1800-12 external integrating sphere, and the spectroradiometer is ASD Fieldspec FR2500. Though the spectral response ranges from 350 nm to 2500 nm, limited by the properties of the light quality of integrating sphere, only the data ranges from 400 nrn to 950 nm were used to analyze. All the spectrum data were sorted to class A （healthy and damaged cotton leaves） and class B （healthy and infected but not damaged cotton leaves）. Because the reflective spectrum of one object is not a curve but a strip, so we can assume that if some object＇s reflective spectra fall into the strip of another object then they are the same object; Otherwise, they are different. According to the figure of class A and class B, we can obviously find that the healthy cotton leaves＇ spectrum concentrate to a thin strip, so healthy cotton leaves＇ spectrum strip was used as a filter to distinguish class A and class B. The classification accuracy of class A and class B was 100% and 92 %, respectively. Classification accuracy was defined as the ratio of samples correctly classified to total samples. In practical application, Land- sat-5 TM remote sensing images multispectral data, which have four bands among 400 nm to 900 nrn are mostly used, so we translated the highspectral data to landsat-5 multispectral data by averaging the value of each banffs spectral range. Also using healthy cotton leaves＇ spectrum strip （Landsat-5） as filter to distinguish class A and class B, the classification accuracies were 96% and 92%, respectively. The results suggest the possibility of using multispectral remote sensing for the survey of cotton disease impacts and merit further study. However, because the high quality reflective spectral data acquired from spectral-radio-integrating sphere avoided the shortcomings caused by natural light and background noise, so in practical application, it is hard to acquire the data above, and the distinguishing accuracy will be lower.