Estimation of apple firmness using hyperspectral spectral indices

Non-destructive and rapid methods for estimating apple firmness are helpful to determine harvesting dates. However, few studies have used an approach of spectral indices for monitoring apple firmness from ground-based handheld hyperspectral data. Therefore, this study analyzed the performance of spectral indices in determining apple firmness based on the spectra of unpeeled apples. The reflectance spectra of apple samples were acquired at wavelength 350-2500 nm. Ten apple varieties were collected from orchards located in seven provinces of China. A two-band combination method (normalized, ratio, and difference type) was used to identify the effective spectral indices. Results showed that the newly developed two-band normalized difference spectral indices (R-840 - R-841)/(R-840 + R-841) (R-lambda is the reflectance value at wavelength lambda) was recommended as the optimal index for monitoring apple firmness, generated the coefficients of determination, root mean square error and residual prediction deviation values between the measured and predicted values of 0.6721, 0.886 kg/cm(2), and 1.7, respectively. This study indicated that spectral indices offer low-cost and portable detection of apple fruit firmness.