【作者】 孙启花；

【导师】 刘向东；

【作者基本信息】 南京农业大学 ， 农业昆虫与害虫防治， 2010， 硕士

【摘要】 农作物病虫害是农业生产上的重要制约因素之一。据联合国粮农组织(FAO)的调查,全世界病虫害每年造成的经济损失达1200亿美元。做好病虫害的准确监测与预测,才能指导病虫的有效防控,从而降低经济损失。遥感技术可以客观、准确、及时的获取地面作物生态环境和作物生长的重要信息,从而间接地对作物病虫害进行监测。地面高光谱遥感是一种无损害快速检测技术,利用它来进行病虫害的监测,可以在一定程度上弥补传统目测手查方法“费时、费力,且准确性受专业水平影响”等的不足。因此,利用遥感技术开展病虫害监测与预警是当今农业遥感的重要发展方向。本文以水稻上的两大重要害虫褐飞虱和稻纵卷叶螟为对象,采用便携式光谱仪,研究了不同生育期水稻、稻田其它绿色植物及非生物地物等的光谱特征；褐飞虱和稻纵卷叶螟不同为害水平下水稻的冠层和叶片光谱特征；同时组建了基于光谱特征的褐飞虱虫量及稻纵卷叶螟为害程度的诊断模型,获得了以下主要结果：(1)以水稻为重点,对不同地物的反射光谱测定结果表明,健康水稻的光谱曲线存在明显的“峰”和“谷”特征,可以显著区别于枯草、水泥地、水塘以及裸土等其它非绿色植物地物。健康水稻光谱曲线在绿光(520-600nm)区有弱反射,出现一个小的峰值；在680-750nm区域内,反射率急剧上升；在700-1000nm处,光谱反射率非常高。枯草的光谱曲线不存在明显的“峰”和“谷”的特征,从可见光到近红外波段反射率逐渐升高,近似于一条直线。水泥地光谱曲线在可见光波段高于其它地物,近红外波段明显低于水稻。水塘的光谱曲线近似于一条低矮直线。裸露土壤的反射率曲线与水稻的在形状上最为相似,但在可见光波段要高于水稻,在近红外波段明显低于水稻。光谱反射率一阶微分能够更清晰地展现地物光谱的变化特征。水稻的一阶微分光谱存在很明显的绿峰(510-560nm)和红边(680-760nm)；而枯草、水泥地、水塘和裸土的微分光谱均表现为一条低矮曲线,没有绿峰和红边,或者绿峰和红边不明显。对孕穗期水稻、抽雄期玉米、结荚期大豆、浮萍和葎草的反射光谱测定分析表明,不同植物间光谱反射率的大小和峰谷位置存在明显差异。表现为,在近红外波段,大豆、玉米和葎草的反射率较高,水稻的次之,而浮萍的最低。在可见光波段,浮萍的反射率最高,玉米、葎草和大豆的次之,而水稻的最低。同一生育期,籼稻的光谱反射率略高于粳稻。对水稻不同生育期的光谱测定发现,在抽穗前,随发育期推进,冠层光谱反射率在可见光范围逐渐降低,近红外波段逐渐升高；但在抽穗以后,随生育期推进,光谱反射率在可见光范围内逐渐升高,而在近红外波段逐渐降低；并且随水稻生育期的推移,冠层光谱的红边呈现出了先“红移”后“蓝移”的规律。比值植被指数RVI和归一化植被指数NDVI也表现为,从移栽期到抽穗期不断增大,而从抽穗期到黄熟期则不断减小的趋势。冠层光谱反射率及植被指数不仅能将水稻与其它绿色植物区分开来,而且可以将不同类型及生育期的水稻区分开来。(2)在室内用水稻标准苗进行了褐飞虱为害后的光谱特性研究,结果表明,在可见光和近红外区,随褐飞虱虫量和虫龄的增加稻株光谱反射率呈下降趋势,近红外光区的光谱反射率可很好地表征褐飞虱不同龄期、不同虫量及成虫产卵对稻株的危害程度。各波长处的光谱反射率与褐飞虱虫量间存在明显的负相关,并且在520-570nm和700-1000nm波长范围内的相关性达到了极显著水平。褐飞虱为害后稻株光谱的红边幅值Dλr和红边面积Sλr也分别与虫量存在极显著相关。利用虫害后稻株在可见光波长550nm处的反射率(R550)和近红外光区波长760nm处的反射率(R760)、红边参数值与没受害稻株相应值的比值建立了褐飞虱虫量的预测模型,利用模型对19个随机虫量进行光谱反演预测,各模型的预测正确率为53%-79%,且R760因子对褐飞虱虫量有相对较好的预测效果,而以5个虫量梯度为一组的5组次实测值与预测值的卡方检测,符合率达80-100%。(3)在水稻孕穗期人工接种不同数量的褐飞虱,然后在灌浆期和黄熟期分别测定水稻的冠层光谱。结果表明,在接虫后20天的灌浆期,水稻冠层光谱在760-1000nm的近红外波段随接虫量的增加而显著降低。在可见光波段,也有下降趋势,但降幅较小。根据冠层光谱反射率与接虫量间的相关性分析得出,725-1000nm波段可做为水稻灌浆期监测褐飞虱的敏感波段。在水稻黄熟期,随接虫量的增加,光谱反射率在可见光区随虫量的升高而显著升高,尤其在620-720nm的红光区上升趋势和幅度非常明显。610-700nm的红光区可作为水稻黄熟期褐飞虱监测的敏感光谱波段。对分蘖期盆栽水稻受不同虫量褐飞虱为害后的冠层、茎部以及叶片光谱反射率的研究发现,冠层的光谱反射率随虫量的增加,在绿光区和近红外光区降低,红光区和蓝光区升高。726-1000nm波段可作为分蘖期水稻冠层褐飞虱监测的敏感波段。单叶和茎部的光谱反射率则表现为在整个波段内均随虫量的增加而降低。剑叶、倒二叶和茎部监测褐飞虱为害的敏感光波区域分别为471-589nm和660-1000nm、489-563nm和708-1000nm、527-569nm和699-1000nm。褐飞虱为害后的水稻光谱植被指数研究表明,冠层差值植被指数DVI、比值植被指数RVI和归一化植被指数NDVI随接虫量的增加而降低,均与接虫量有较好的相关性。叶片光谱和茎部光谱的DVI与接虫量的相关性显著,而RVI和NDVI与虫量之间相关性不显著。RVI和NDVI指数可用于冠层光谱层次上的褐飞虱监测,而不适宜于单叶和茎杆光谱层次上的监测。孕穗期褐飞虱的接虫量大小与水稻的产量关系密切,随着接虫量的增多,水稻穗重和千粒重呈显著的下降趋势。水稻黄熟期的光谱参数：R696、Dλy、λr、Dλr和Sλr与水稻产量间存在显著的相关性。建立了褐飞虱为害后基于水稻黄熟期光谱参数的水稻产量估算模型,估测千粒重(Y3)的两个模型Y3=0.143λr-99.272和Y3=383.121Dλr+1.931的估产效果较好,可用于监测不同虫量为害下水稻的产量。(4)测定了稻纵卷叶螟为害后水稻的冠层光谱,并在稻纵卷叶螟为害的小区内选取受害叶片与没受害叶片,在室内测定了单叶光谱,结果表明,稻纵卷叶螟为害后的水稻,在冠层水平上随着卷叶率的升高,反射率在近红外光区逐渐降低。冠层水平上的敏感波段为：738-1000nm。在已受害的卷叶水平,随小区内卷叶率的升高,反射率在红光波段逐渐升高,表征稻纵卷叶螟为害的敏感波段为582-688nm的红光波段。在没受害的展开叶水平上,随小区内卷叶率的升高,反射率在可见光区和近红外光区均降低,表征稻纵卷叶螟为害的敏感波段为512-606nm的绿光区和699-1000nm近红外光区。水稻冠层光谱反射率的一阶微分结果表明,红边位置λr随卷叶率的加重而“蓝移”,红边幅值Dλr和红边面积Sλr均随受害等级的加重而减小。Dλr和Sλr与水稻受害等级存在极显著的相关性,该光谱指数可用于监测水稻受稻纵卷叶螟为害的程度。建立了基于光谱特性的稻纵卷叶螟为害程度的预测模型,冠层光谱条件下,以Y=-1338.406Dλr+11.123模型的效果最好；没直接受害的展开叶光谱条件下,以Y=-90.280R550+11.902模型的效果最好；直接受害的卷叶光谱条件下的预测模型为Y=198.620R670-18.505,所建模型均通过了0.01水平的检验,基本可指导虫害监测。更多还原

【Abstract】 Insect and disease pests are one the factors to restrict the development of agricultural production. According to the survey of FAO, the worldwide economic losses caused by insect and disease pests reached to 120 billion dollars every year. Accurate monitoring and forecasting can guide the effective management and control of insect and disease pests, and reduce the economic losses. Remote sensing technology can objectively, accurately, and duly get the information of the ground crop’s ecological environment and growth conditions, and indirectly monitor the number and damage degree of the insect and disease pests. Traditionally, assessment and monitoring of diseases and insect pests in plants is being done by visual approach, relying upon the human eye and brain, which is time-consuming, labour intensive, and the accuracy affected by the professional level of observer. Hyperspectral remote sensing is a fast and lossless technique, which can make up the shortcoming of the traditional techniques. Therefore, using remote sensing technology to monitor and forecast insect and disease pests is an important development direction of agricultural remote sensing.In this paper, we focus on the brown planthopper Nilaparvata lugens (Stal) (BPH) and rice leaf roller Cnaphalocrocis medinalis (Guenee) (RLR), and use the hyperspectral remote sensing, to study on the spectral characteristics of rice at different growth stages, the other plants in the rice field and non-biological ground objects. The spectrum of leaf and canopy of rice infested by BPH and RLR was also measured and the monitoring models to forecast the number of BPH and the damage level of RLR were established based on the spectral characteristics. The results are as follows:The results of reflectance features of different ground objects showed that there are obvious "peak" and "valley" features of the spectral curve of healthy rice, which is significantly different from the dead grass, concrete ground, pool, bare soil and other non-green plants. The spectral curve of healthy rice has a small reflection peaks in the green wave band (520-600nm). The reflectance increased sharply in the 680-750nm region, and it reached up to the maximum in 700-1000nm. There were no obvious "peak" and "valley" features in the dead grass spectral curve, and it gradually increased as a smooth straight line from visible band to near-infrared band. The spectral curve of concrete ground was the highest than that of the other objects in the visible band, but it was significantly lower than rice in the near-infrared band. The spectrum of the pool was approximated to a low level of linear. Reflectance curve of the bare soil was similar to rice in the shape, but the reflectances in the visible band were higher than these of rice, while significantly lower than rice in the near-infrared band. The first derivative of spectral reflectance could exhibit more clearly the changes in spectral characteristics of the different ground objects. There was a very obvious green peak at 510-560nm and a red edge at 680-760nm in the first derivative of spectral reflectance of rice, but the first derivative of spectra of all these ground objects, dead grass, concrete ground, pool and bare soil only showed a low curve, and no or not obvious green peak and red edge.The spectral reflectance analysis of the rice on booting stage, heading stage corn, pod of soybean, duckweed and humulus scandens, indicated that the spectral reflectance and location of the peak and valley among the different plants were significantly different. The spectral reflectance of soybean, corn and humulus scandens were the highest at the near-infrared band, followed by rice, but duckweed was the lowest. In visible band, the spectral reflectance of duckweed was the highest, followed by corn and humulus scandens, while rice was the lowest. The spectral reflectance of the indica rice was slightly higher than that of the japonica rice at the same growth period.Spectral characteristics of rice at different growth stages showed that before the heading stage of rice, the spectral reflectance increased gradually at visible band with the advance of the growth stage, and decreased in the near infrared band. But after the heading stage, spectral reflectance decreased gradually at visible band, and increased at the near-infrared band with the advance of the growth stage. The red edge location of the rice canopy moved to the longer band called "Red transference", and then moved to the shorter band called "Blue transference" as the growth stages of rice developed. Ratio vegetation index RVI and normalized difference vegetation index NDVI of rice increased from transplanting stage to heading stage, and decreased from heading stage to yellow ripeness stage. Canopy spectral reflectance and vegetation index of rice could distinguish well not only the rice from the other plants, but also the different types and growth stages of rice.The spectrum of rice seedling infested by brown planthopper was measured in the laboratory. The results showed that the spectral reflectances at the range of visible light and near infrared regions decreased significantly with the increase of the number and instar of BPHs. The damage degrees of rice plants caused by the BPH nymphae with different numbers or stars, and by the oviposition behaviour of adult were expressed well by the spectral reflectance in the near-infrared wavelengths. The reflectance was negatively correlated with the number of BPHs, and the correlation coefficients were significant at the range of wavelengths of 520-570nm and 700-1000nm. The red edge slope (Dλr) and red edge area (Sλr) of the reflectance also significantly correlated with the number of nymphae. The linear models for forecasting the occurrence number of BPHs were built using the following relative indexes to the undamaged plants:the spectral reflectance in the wavelengths of 550nm (R550) and 760nm (R760), and the red edge indexes (Dλr and Sλr). The accuracy of the models was 53%-79% for the 19 times tests. The factor of R760 was efficient for forecasting the number of BPHs. The chi-square test demonstrated that the coincidence rate between the real values of the 5 series of number of BPH and the calculated numbers by models achieved 80%-100%.The canopy spectral characters of rice were measured and analyzed in the filling stage and yellow ripeness stage, respectively, after being inoculated different amounts of BPH at the booting stage. The results showed that in the filling stage, canopy spectral reflectance significantly decreased at 760-1000nm of near-infrared region with the increase of BPH, and there was also a decreased trend in the visible band. The correlation between canopy reflectance and the number of BPH showed that 725-1000nm was the sensitive wave brands for monitoring BPH in the filling stage. In the yellow ripeness stage, the canopy reflectance of rice increased in the visible region as the number of BPH increased, especially in the red band region (620-720nm). So the wave brands 610-700nm could be considered as the sensitive band to monitor the number of BPH at the yellow ripeness stage of rice.Spectral reflectances of the canopy, stem and single leaf of rice plant infested by various numbers of BPH were measured at the tillering stage. Results showed that canopy reflectance decreased at the green and near-infrared region, while it increased at blue and red bands with the increase of BPH. The sensitive wave band to BPH under the level of canopy was 729-1000nm. The spectral reflectances of the single leaf and stem in the whole bands were decreased with the increase of the number of BPH. According to the results of the correlative analysis between reflectance and the number of BPH, the sensitive wave bands to BPH under the level of flag leaf, next-to-last and stem were 471-589nm and 660-100nm,489-563nm and 708-100nm,527-569nm and 699-1000nm, respectively.The spectral vegetation index of the rice showed that canopy spectrum DVI, RVI, and NDVI decreased significantly with the increase of BPH. All of these indexes had a significant correlation with the number of BPH. The spectrum DVI index of leaves and stems was significantly correlated to the amount of BPH, but there was no significant correlation between the number of BPH and RVI or NDVI. Spectral vegetation index RVI and NDVI would be used to monitor the BPH on the canopy level, but they were unsuitable for the single leaf and stem level.The relationship between the rice yield and the number of BPH at the booting stage of rice was very significant. The weight of a spike and 1000-seed of rice significantly decreased with the increase of BPH. There was significantly correlation between rice yield and the spectral parameters (R696, Dλy,μr, Dμr and Sμr) in yellow ripeness stage. The models to assess the yield of rice were established based on the spectral parameters. The two modelsY3=0.143kr-99.272 and Y3=383.121Dμr+1.931 were efficient to monitor the 1000-seed weight (Y3) of rice under different amount of BPHs.Canopy and single leaf reflectance of rice infested by RLR were measured. The results indicated that the spectral reflectance of rice canopy decreased in the near-infrared regions with the increase of the damage degree by RLR.738-1000nm was the sensitive wave-length band which could reflect the impaired level of rice canopy. The spectral reflectance of the roll leaf increased in the red band regions with the increase of the rate of roll leaf in the plot of rice, and 582-688nm was the sensitive band to reflect the impaired level of rice leaf-roll. The spectral reflectance of the developed blade leaf in the damaged plot by RLR decreased in the visible band and near-infrared regions with the increase of the rate of the roll leaf. The sensitive wave bands to RLR under the level of developed blade leaf were 512-606nm and 699-1000nm. First derivative of canopy spectral reflectance showed that location of the red edge (λr) moved to the short band side. Dλr and Sλr decreased with the increase of infestation level of leaves. The Dλr and Sλr were significantly correlative to the infestation level of rice. These two indexes could be used to detect the damage degree of rice by RLR. Prediction models based on the spectral characteristics of rice were established to forecast the damage levels by RLR. At the canopy level, Y=-1338.406DAλr+11.123 was the best model, whereas at the developed blade leaf level, Y=-90.280R550+11.902 was the best one, and the model was Y=198.620R670-18.505 at the damaged leaf-roll leaf level. All these models were tested through the 0.01 level, and they could be used to monitor the RLR.更多还原