Estimation Of The Growth For Radish And Chinese Cabbage Using Hyperspectral Image

Abstract

Background: Hyperspectral camera is useful for finding a wavelength range that can estimate growth by using wide spectrum (400-1000nm) ranges. It is possible to apply precision agriculture through determination of accurate harvest season, amount of applied fertilizer and so on according to estimation of growth. The objective of this study was to estimate the growth for radish by using hyperspectral image depending on vegetation stages. Methods: Hyperspectral images of Radish and Chinese cabbage were acquired at intervals of 2 weeks in midday. It was divided into the crop and background area in image by using NDVI and the reflectance values of crop area were calibrated by the values of reference board in same image. PLSR analysis was employed to develop model for estimating growth of the crops with the all spectral band of all vegetation stages. The models for estimating the weight of the crops were evaluated through R 2 and root mean square error (RMSE), which were verified as validation model by full-cross validation. Results: The model for estimating the fresh weight for Radish exhibited high performance (R2=0.950, RMSE=236.3g), but validation model exhibited high error value (R2=0.806, RMSE=492.3g). Likewise, the model for estimating the dry weight exhibited high performance (R2=0.975, RMSE=3.413g), but validation model exhibited high RMSE (R2=0.843, RMSE=8.993g). The model for estimating the fresh weight for Chinese cabbage exhibited high R2 (R2=0.897, RMSE=571.1g), but estimation and validation model exhibited high error value (R2=0.859, RMSE=685.0g). Discussions: The estimation and validation model of the fresh and dry weight for the crops are required to improve RMSE by using different statistics analysis and various vegetation indices. Also, comparative analysis between spectral attributes of this experiment and repeated experiment are required. Conclusion: It is possible to mostly estimate growth of the crops, but estimation and validation models of weight are needed to improve RMSE.