Rapid estimation of seed yield using hyperspectral images of oilseed rape leaves

In this study we developed a technique to early and rapidly estimate seed yield using hyperspectral images of oilseed rape leaves in the visible and near infrared (VIS–NIR) region (380–1030 nm). Hyperspectral images of leaves were acquired four times from field trials in China between seedling until pods stage. Seed yield data on individual oilseed rape plants were collected during the local harvest season in 2011. Partial least square regression (PLSR) was applied to relate the average spectral data to the corresponding actual yield. We compared four PLSR models from four growing stages. The best fit model with the highest coefficients of determination (RP2) of 0.71 and the lowest root mean square errors (RMSEP) of 23.96 was obtained based on the hyperspectral images from the flowering stage (on March 25, 2011). The loading weights of this resulting PLSR model were used to identify the most important wavelengths and to reduce the high dimensionality of the hyperspectral data. The new PLSR model using the most relevant wavelengths (543, 686, 718, 741, 824 and 994 nm) performed well (RP2=0.71, RMSEP = 23.72) for predicting seed weights of individual plants. These results demonstrated that hyperspectral imaging system is promising to predict the seed yield in oilseed rape based on its leaves in early growing stage.

Figure optionsDownload as PowerPoint slideHighlights
► We analyzed hyperspectral images of oilseed rape leaves for early yield prediction.
► PLSR model gave a good correlation between seed yield and leaf spectra.
► Optimum wavelengths were selected to reduce dimensionality in hyperspectral analysis.