The objects' spectrum is often contaminated by noise when it is collected in the open air. According to the principle of the spectrum collection, the noise was considered as one kind of multiplicative compound noise. By theoretical derivation, the combination of logarithm transform and wavelet transform was introduced into noise reduction. Multiplicative noise simulation test was carried out. And the results show that the spatial correlation algorithm is best suited for spectral data denoising, modulus maxima algorithm is inferior to it. Threshold shrinking rule is unsuitable for spectrum denoising. The wild plants spectrum were processed based on spatial correlation algorithm. Results show that the noise near 1450 nm in the spectrum is perfectly denoised, while near 1800 ～ 1900 nm strong noise can not be removed perfectly. The reason is the limited records accuracy of the spectrometer. When the theoretical ratio is far greater than 1, the spectrometer will accurately record them. While the theoretical ratio is far less than 1, the record will be 0. So serious system errors will be generated in strong noise band and will be retained after the wavelet transform was applied because they are considered as signal singularity. Experiments prove that spatial correlative filtering with the combination of logarithm transform and wavelet transform is feasible for multiplicative-noise-contaminated spectrum denoising