Improvement of hyperspectral imaging signal quality using filtering technique

This paper proposes to improve signal quality in hyperspectral imaging (HSI) on the basis of noise analysis and filtering method. HSI technology enables nondestructive and precise analysis in agriculture and food industries by acquiring high-resolution images over multiple wavelengths, but the identification and removal of noise in the signal is a challenge. In this study, HSI measurement data of sucrose solution samples of different concentrations were used as experimental subjects. The outliers outside the three-fold standard deviation range of all data were identified as noise and a filtering method using noise mask and Wavelet transform was proposed. By evaluating the effect of the filtering method on noise reduction, we conducted qualitative and quantitative analysis and comparison, mainly through statistical methods and the limits of detection (LOD), LODmin and LODmax. The experiment results show that the proposed method is useful in removing noise, reducing the detection limit when applying Partial Least Squares (PLS) and improving the HSI signal quality. This is expected to improve the accuracy of nondestructive analysis using HSI data. • Noise can be detected due to the normal distribution of each HSI wavelength data. • Outliers outside three-fold standard deviation range were identified as noise. • We proposed Filtering method using mask, neighbor pixels and Wavelet transform. • Proposed method significantly reduced the outlier rate and limits of detection. • Proposed method better suited than existing methods for removing outlier noise.