Vegetation field spectrum denoising via lifting wavelet transform

Abstract

Field spectrum pretreatment experiments were carried out, and denoising numerical experiment via lifting wavelet transform (LWT) was designed, and several famous test signals including blocks, bumps, heavy sine and doppler were processed via LWT in these experiment. And the field spectrum was processed via LWT. Experiments proved that S NRG-to-S NRN curves have similar feature and they all have a peak. And SNRG of almost all employed wavelets have higher value with S NRN between 0 and 20 dB. When signal is at high S NR, the SNRG is very little, and the M SED of denoised signal became little by little. LWT is more suite to denoise the low S NR or heavy noise contaminated signals. Bior4.4 have wider S NRN interval for denoising comparing with other five wavelets, including haar, db6, sym6, bior2.2 and bior3.3. Original field spectrum is processed by 3 stage liftings based on bior4.4 to denoise the trivial noise-contaminated regions. On processing the water band signal, logarithm transform is firstly taken. And then the spectrum is denoised via LWT based on bior4.4. The results show that an excellent denoised spectrum can be get, especially between 350 nm and 1 800 nm, and between 1 960 nm to 2 500 nm. While there is still a bump around 1 900 nm, this maybe due to the spectrum machine’s limited precision.