野外植物波谱中噪声分布规律研究

Noise Analysis on Field Collected Vegetation Spectrum

基于空间相关算法和小波方差估计法,进行了噪声检测.实验表明,野外波谱噪声总体方差数量级介于10^-2-100之间,在350^-1- 300 nm,野外波谱曲线的噪声方差略高于JPL光谱库中的Conifers,1 450 nm和1 800 nm附近则噪声相当显著.野外波谱曲线的小波系数能量分布与加性仿真信号不同.在1 450 nm和1 800 nm附近,噪声在高分解尺度上表现为尖锐毛刺的减少,噪声能量减弱;同时在这2个区段的信号总体能量逐渐增大,证明有波谱特征信号存在.野外波谱数据中这种尖锐噪声与2个污噪信号相除得到的复合信号表现出相似的特征,其能量传播特征也相似.

In this article noise existing in field collected vegetation spectrum was analyzed via wavelet spatial algorithm. Bumps, heavy sine and vegetation spectrum from JPL were used as noise contaminated signals and simulating experiments verified its effectiveness. Noise detecting experiments proved that heavy noise existed in field spectrums; the total variance was up to 10^-2 to 10^0 and far exceeded that from the JPL. Comparing with the spectrums from JPL, noise mainly existed nearby 1 450 nm and 1 800 nm. The field collected spectrums＇ wavelet coefficients energy distribution was different to that of the additive noise contaminated signal. Nearby 1 450 nm and 1 800 nm, incisive burr vanished in higher decomposition scale, the noise energy was getting more and more weak; Meanwhile signal＇s overall energy increased gradually in these two sector＇s. Incisive noise and its energy dissemination characteristic in field collected spectrum were similar to that of two dividing heavy contaminated signals by random noises. The theoretical analysis proved that noise in the spectrum curvels is one kind of productive compound noise. This kind of noise was difficult to remove by threshold denoising method. Noise elimination method can be designed according to the noise distribution.