Development of data reduction function for hyperspectral imaging

Abstract

Algorithms have been developed for controlling the calibration and the measurement cycles of hyperspectral equipment. Special calibration and preprocessing methods were necessary to obtain suitable signal level and acceptable repeatability of the measurements. Therefore, the effect of the noise of NIR sensor was decreased, the signal level was enhanced and stability was ensured simultaneously. In order to investigate the properties of the number of objects suitable for statistical analysis, the enormous size of acquired hypercube (gigabytes per object) should be reduced by vector-to-scalar mathematical operators in real-time to extract the desired features. The algorithm developed was able to calculate the score of operators during scanning and the matrices were displayed as pseudo-images to show the distribution of the properties on the surface. The operators had to be determined by analysis of a sample set in preliminary experiments. Stored carrot was chosen as a model sample for investigation of the detection of moisture loss by hyperspectral properties. Determination of the proper operator on different tissues could help to analyze and model drying process and to control storage. Hyperspectral data of different carrot cultivars were tested under different storage conditions. Using improved measurement method the spectral parameter of the suitable operator described quite well the moisture loss of the different carrot tissues.