Optical Imaging For Rapid Detection Of Citrus Huanglongbing

Huanglongbing(HLB)is a highly contagious and devastating citrus disease,however no effective treatments have been found until now.It has a long latent period,and the time from infection to typical symptoms appearance varies from several months to years depending on tree age,citrus cultivars and environmental conditions.Infected trees can act as inoculum sources in the asymptomatic period,resulting in a rapid spread of HLB disease in the orchard.Therefore,it is necessary to remove the infected trees as early as possible.Optical imaging techniques are promising for rapid and non-destructive detection of plant diseases.Hyperspectral images and chlorophyll fluorescence images of citrus leaves in different seasons,orchards and infected stages were collected,and the starch,sucrose,glucose and fructose were also measured.The aim of this study was to establish a rapid detection method for HLB disease using optical imaging,and elucidate the effects of seasonal changes,different orchards and the infected degrees on the host carbohydrate metabolism,photosynthesis and the performance of HLB detection.The contents and conclusions of this study are listed as follows:Firstly,dynamics of carbohydrate metabolism of HLB infected citrus leaves under different seasons,orchards and infection stages were investigated.The abnormal metabolism of carbohydrate was already observed in asymptomatic HLB-infected leaves.It was also found that accumulation of carbohydrates began in asymptomatic period,and sucrose and glucose in HLB-infected leaves occurred earlier than that of starch.Although nutrient-deficient(Fe)and HLB-infected leaves showed similar symptom,however,carbohydrates content in nutrient(Fe)deficient leaves presented a reverse pattern to HLB-infected ones.Secondly,changes of visible and near infrared spectral reflectance of HLB-infected leaves were studied.The blockage of phloem and accumulation of carbohydrates in HLB-infected leaves can damage the pigment and structure of leaves which in turn can change the light properties.The reflectance of HLB-infected and nutrient(Fe)deficient leaves increased in the visible range.A higher reflectance in the near-infrared region of HLB-infected leaves was observed,but Fe-deficient leaves showed an opposite pattern.Five optimal wavelengths(493,515,665,716 and 739 nm)related to the yellowing symptom and the biochemical composition were finally selected to describe the spatial and spectral features of HLB disease in different seasons,orchards and infection stages.Classification results from least squares-support vector machine(LS-SVM)model demonstrated a good linkage between the spectral and the textural features and HLB fingerprint.Overall accuracies of 91.6%and 89%were obtained for symptomatic leaves in Orchard 1 and asymptomatic HLB-infected leaves in Orchard 2,respectively.Additionally,the robustness of the detection model was successfully evaluated by a different citrus cultivar(Ponkan)using the model transfer strategy with overall accuracy of 93.5%.Thirdly,photosynthetic response of citrus leaves to HLB infection was also analyzed.HLB infection already caused irreversible damage to the photosynthetic system in citrus leaves with an increase of minimum fluorescence(Fo),decrease of maximum quantum yield of PSII(Fv/Fm)and the number of active light reaction centers(Fv/Fo).The quantum yield of PSII(?PSII)HLB-infected leaves was reduced,indicating a decreasing ability of converting energy fluxes into photochemistry of host.However,the energy fluxes allocated as non-regulated energy dissipation for photoinhibition(?NO)increased due to an irreversible damage to PSII center,and this damage even happened at the asymptomatic stage.The feasibility of chlorophyll fluorescence imaging for HLB disease detection in different seasons,orchards and infection stages was also evaluated.It was found that chlorophyll fluorescence signatures derived from protocol L4 could provide more photosynthetic fingerings about HLB disease.It achieved more than 95%and 89%overall accuracy throughout the whole experimental period for Orchard 1 and Orchard 2(except for new emerging leaves in May and September),respectively,using LS-SVM model.It demonstrated that 53 chlorophyll fluorescence signatures derived from protocol L4 were more suitable to describe photosynthetic features of HLB-infected leaves in different seasons,orchards and infection stages.Lastly,the robust citrus HLB detection model was established by fusing hyperspectral imaging and chlorophyll fluorescence imaging.The visible/near-infrared hyperspectral images and chlorophyll fluorescence images are strongly related to the leaf pigment,biochemical composition and cell structure,and photosynthesis of citrus leaves.It is possible to perform sensor fusion to obtain HLB 'fingerprints' with a higher reliability and spatial-temporal resolution.Results have shown that the LS-SVM model based on combination of five optimal wavelengths(493 nm,515 nm,665 nm,716 nm and 739 nm)and 29 chlorophyll fluorescence signatures derived from protocol L2 could improve discriminant ability.It can reduce the false negative rate(11.6%and 17.6%,respectively,for May and September in Orchard 2),computational time(56.7%)and the number of model inputs(35.8%).It achieved 96.7%and 89.5%detection accuracies for Orchard 1 and Orchard 2,respectively,which was better than either using spectral reflectance at optimal wavelengths(85.9%and 79.4%for Orchard 1 and Orchard 2,respectively)or 29 chlorophyll fluorescence signatures from protocol L2(95.5%and 84.6%for Orchard 1 and Orchard 2,respectively).The findings in this research demonstrated the feasibility of fusing hyperspectral imaging and chlorophyll fluorescence imaging for HLB disease detection in different seasons,orchards and infected stages.