Wavelength selection in hyperspectral imaging for prediction banana fruit quality

Wavelength selection is one of the most problems in hyperspectral image analysis due to an immense amount of hypercube data. In this paper, a method to find the optimal wavelength selection in predicting the quality of the banana fruit (Musa sp.) was presented. Reduction of the dimension of wavelength was performed in two stages. The peak and valley detection of reflectance profile from sample data was conducted to find a candidate of the optimal band. The modified of CARS method was then used to select the best wavelength using the Standard deviation information. The PLS registration then used to compute the predicted value based on the transfer model that created using training data. The squared relative errors and correlation coefficient was selected to compare the performance of selected wavelength. The best performance is provided by modified CARS method which has 15 optimal wavelength band with the square of the relative error of 0.09 and the correlation coefficient R
²
of 0.97 in predicting the total soluble solid content in banana.