Application and Evaluation of Wavelet-Based Denoising Method in Hyperspectral Imagery Data

The imaging hyper-spectrometer is highly susceptible to the presence of noise and its noise removal is regularly necessary before any derivative analysis. A wavelet-based(WT) method is developed to remove noise of hyperspectral imagery data, and commonly used denoising methods such as Savitzky-Golay method(SG), moving average method(MA), and median filter method(MF) are compared with it. Smoothing index(SI) and comprehensive evaluation indicator(η) are designed to evaluate the performance of the four denoising methods quantitatively. The study is based on hyperspectral data of wheat leaves, collected by Pushbroom Imaging Spectrometer (PIS) and ASD Fieldspec-FR2500 (ASD) in the key growth periods. According to SI andη, the denoising performance of the four methods shows that WT>SG=MA>MF and WT>MA>MF>SG, respectively. The comparison results reveal that WT works much better than the others with the SI value 0.28 andηvalue 5.74E-05. So the wavelet-based method proposed in this paper is an optimal choice to filter the noise, in terms of balancing the contradiction between the smoothing and feature reservation ability.