Classification Of Potato Varieties Drought Tolerance Based On Hyperspectral Imaging And Screening Of Drought-tolerant Potato Varieties

In recent years,hyperspectral technology has been widely used in agricultural production with its characteristics of non-destructive and high-throughput,which can be used for plant nutrition status judgment,plant growth monitoring,yield detection and estimation,early warning and control of diseases and pests,etc.Hyperspectral technology is also used in potato related research,often for phenotypic monitoring,such as potato plant height extraction,aboveground biomass estimation,etc.,and physiological index monitoring,but hyperspectral technology has not been systematically reported in potato tolerance classification studies.,In this study,hyperspectral imaging was used to classify and predict potato drought grade,and the reflectance of potato leaf spectrum under different drought treatments was comparatively analyzed.Different algorithms such as multivariate scattering correction(MSC)and SavitzkyGolay convolution smoothing(SG)were preprocessed,and then SPA and CARS were used to extract the characteristic wavelengths.Through the accuracy evaluation and different modeling methods,the optimal model for the classification and identification of drought tolerance grades of potato leaves was finally established.In this experiment,the drought level of potato leaves was visualized.At the same time,we screened a number of different potato varieties from agronomic traits,yield indicators and quality indicators under drought conditions,and the main results were as follows:(1)The spectral index of some potato leaves under different drought stress treatments was different,but the drought tolerance grade could not be classified by the spectral index.(2)The spectral reflectance of potato leaves was typed with the extension of drought stress time,and the accuracy of drought tolerance grade classification model gradually increased.(3)The preprocessing results of the spectral reflectance of potato leaves showed that the optimal methods were MSC and SG,and this method could give full play to the maximum performance of the classification model.(4)The results of the selected number of feature bands are classified,and the results showed that the CARS is optimal,and the method that can maximize the performance of the model is established as the independent variable extraction method of MSC-CARS.(5)By comparing the accuracy of the characteristic bands under different treatments of the CARS method,it is found that the optimal model is MSC-CARS-EXC,the accuracy is94.12%.And the optimal parameters were established.(6)The optimal classification model for selecting characteristic wavelengths is CARSGBDT,and the visualization of drought grade classification results shows that with the increase of drought level,the color of the image will also change,from blue(normal water)to green(moderate drought stress)to yellow(severe drought stress).(7)We screened out three drought-resistant varieties,Longshu No.7,Qingshu No.9and Zhuangshu No.3,which were suitable for planting in Xunyi area.