Rice yellow mottle is considered the most destructive disease threatening rice production in Africa. Early detection of this infection in rice is essential to limit its expansion and proliferation. However, there is n...Rice yellow mottle is considered the most destructive disease threatening rice production in Africa. Early detection of this infection in rice is essential to limit its expansion and proliferation. However, there is no research devoted to the spectral detection of rice yellow mottle virus (RYMV) infection, especially in the asymptomatic or early stages. This work proposes the use of hyperspectral fluorescence and reflectance data at leaf level for the detection of this disease in asymptomatic stages. A greenhouse experiment was therefore conducted to collect hyperspectral fluorescence and reflectance data at different stages of infection. These data allowed to calculate nine vegetation indices: one from fluorescence spectra and eight from reflectance spectra. A t-test made it possible to identify, from the second day after infection, four relevant reflectance vegetation indices to discriminate healthy leaves from those infected: these are Photochemical Reflectance Index (PRI), Transformed Chlorophyll Absorption in Reflectance Index (TCARI), Structure Intensive Pigment Index (SIPI) and Simple Ratio Pigment Index (SRPI). The fluorescence index was less sensitive in detecting infection. The four significant vegetation indices for the detection of RYMV were then used to build and evaluate models for discriminating plants according to their health status by the supervised classification of support vector machine (SVM) at different stages of infection. The maximum overall accuracy is 92.5% six days after inoculation (6 DAI). The sixth day after inoculation would be the adequate day to detect RYMV. This plants discrimination was validated by the mean reflectance spectra and by the histograms showing the differences between the average reflectance vegetation indices values of the two types of plants. Our results demonstrate the feasibility of differentiating RYMV-infected samples. They suggest that support vector machine learning models could be developed to diagnose RYMV-infected plants based on vegetation indices derived from spectral profiles at early stages of disease development.展开更多
文摘Rice yellow mottle is considered the most destructive disease threatening rice production in Africa. Early detection of this infection in rice is essential to limit its expansion and proliferation. However, there is no research devoted to the spectral detection of rice yellow mottle virus (RYMV) infection, especially in the asymptomatic or early stages. This work proposes the use of hyperspectral fluorescence and reflectance data at leaf level for the detection of this disease in asymptomatic stages. A greenhouse experiment was therefore conducted to collect hyperspectral fluorescence and reflectance data at different stages of infection. These data allowed to calculate nine vegetation indices: one from fluorescence spectra and eight from reflectance spectra. A t-test made it possible to identify, from the second day after infection, four relevant reflectance vegetation indices to discriminate healthy leaves from those infected: these are Photochemical Reflectance Index (PRI), Transformed Chlorophyll Absorption in Reflectance Index (TCARI), Structure Intensive Pigment Index (SIPI) and Simple Ratio Pigment Index (SRPI). The fluorescence index was less sensitive in detecting infection. The four significant vegetation indices for the detection of RYMV were then used to build and evaluate models for discriminating plants according to their health status by the supervised classification of support vector machine (SVM) at different stages of infection. The maximum overall accuracy is 92.5% six days after inoculation (6 DAI). The sixth day after inoculation would be the adequate day to detect RYMV. This plants discrimination was validated by the mean reflectance spectra and by the histograms showing the differences between the average reflectance vegetation indices values of the two types of plants. Our results demonstrate the feasibility of differentiating RYMV-infected samples. They suggest that support vector machine learning models could be developed to diagnose RYMV-infected plants based on vegetation indices derived from spectral profiles at early stages of disease development.
